Msc_thesis_report_naman_negi.

Transcript

DEPARTMENT OF PRECISION AND MICROSYSTEMS ENGINEERING DIFFERENCES IN STEERING BEHAVIOUR BETWEEN EXPERTS, EXPERIENCED AND NOVICE DRIVERS: A DRIVING SIMULATOR STUDY NAMAN SINGH NEGI (4180925) REPORT NO AUT 2013.022 COACH DR.IR RIENDER HAPPEE PETER VAN LEEUWEN, MSc PROFESSOR DR.IR EDWARD HOLWEG SPECIALISATION AUTOMOTIVE TYPE OF REPORT MSC THESIS REPORT DATE 24 August 2013 TABLE OF CONTENTS ABSTRACT ...................................................................................................................................................... 6 CHAPTER 1: INTRODUCTION ................................................................................................................... 7 1.1 DRIVING TASK ...................................................................................................................................... 7 1.2 SAFETY AND THE ROLE OF DRIVER ASSIST SYSTEMS ............................................................... 8 1.3 RESEARCH PROBLEM ......................................................................................................................... 9 1.3.1 RESEARCH STATEMENT ................................................................................................................ 9 1.3.2 METHODOLOGY ADOPTED........................................................................................................ 10 1.4 THESIS OUTLINE ................................................................................................................................ 10 CHAPTER 2: LITERATURE STUDY ........................................................................................................ 11 2.1 PREVIOUS RESEARCH....................................................................................................................... 11 2.2 INTELLIGENT ADAS SYSTEMS........................................................................................................ 14 2.3 SUMMARY ........................................................................................................................................... 15 CHAPTER 3: RACE VERSUS EXPERIENCED DRIVERS.................................................................... 17 3.1 INTRODUCTION.................................................................................................................................. 17 3.2 METHODS............................................................................................................................................. 17 3.2.1 APPARATUS ................................................................................................................................... 17 3.2.2 EXPERIMENT INSTRUCTIONS .................................................................................................... 18 3.2.3 PARTICIPANTS .............................................................................................................................. 18 3.2.4 DEPENDANT MEASURES............................................................................................................. 19 3.3 RESULTS............................................................................................................................................... 20 3.3.1. ROAD DEPARTURES.................................................................................................................... 20 3.3.2 CURVE-TIMES ............................................................................................................................... 21 3.3.3 LATERAL ACCELERATION........................................................................................................... 23 3.3.4 STEERING PERFORMANCE......................................................................................................... 25 3.3.5 PATH STRATEGY ........................................................................................................................... 26 3.4 DISCUSSION......................................................................................................................................... 29 CHAPTER 4: NOVICE VERSUS EXPERIENCED DRIVERS ............................................................... 31 4.1 INTRODUCTION.................................................................................................................................. 31 4.2 METHODS............................................................................................................................................. 31 4.2.1 X-CAR SIMULATOR....................................................................................................................... 31 4.2.2 EXPERIMENT INSTRUCTIONS, PART A: DOUBLE LANE CHANGE ....................................... 32 4.2.2 EXPERIMENT INSTRUCTIONS, PART B: HIGH SPEED CORNERING .................................... 33 4.2.3 PARTICIPANTS .............................................................................................................................. 34 4.2.4 DEPENDANT MEASURES............................................................................................................. 34 4.2.5 QUESTIONNAIRE .......................................................................................................................... 36 4.3A RESULTS PART A: DOUBLE LANE CHANGE .............................................................................. 37 4.3A.1 PATH FOLLOWED ...................................................................................................................... 37 4.3A.2 STRATEGY AND PERFORMANCE............................................................................................. 41 2 4.3A.3 CONTROL STRATEGY................................................................................................................. 45 4.3A.4 TEST RE-TEST REPEATABILITY................................................................................................ 48 4.3A.5 DISCUSSION................................................................................................................................ 49 4.3B RESULTS PART B: CORNERING .................................................................................................... 50 4.3B.1 CURVE-TIMES AND LATERAL ACCELERATION..................................................................... 50 4.3B.2 STEERING PERFORMANCE ...................................................................................................... 51 4.3B.3 PATH STRATEGY......................................................................................................................... 53 4.3B.4 DISCUSSION................................................................................................................................ 55 CHAPTER 5: CONCLUSION AND FUTURE RESEARCH.................................................................... 57 5.1 CONCLUSION ...................................................................................................................................... 57 5.2 FUTURE RESEARCH........................................................................................................................... 60 REFERENCES ............................................................................................................................................... 62 APPENDICES ................................................................................................................................................ 64 APPENDIX 1. CONSENT FORM .............................................................................................................. 64 APPENDIX 2: QUESTIONNAIRE............................................................................................................. 66 APPENDIX 3: CRASHES ANALYSIS ...................................................................................................... 68 APPENDIX 4: TLX QUESTIONNAIRE .................................................................................................... 71 APPENDIX 5: VEHICLE DYNAMICS QUESTIONNAIRE .................................................................... 75 APPENDIX 6: MATLAB PROGRAMS ..................................................................................................... 77 TABLE OF FIGURES Figure 1: RTrainer car ...................................................................................................................................... 18 Figure 2: Mallory Park Test Circuit ................................................................................................................. 18 Figure 3: Track breakdown into different curves for analysis ......................................................................... 19 Figure 4: X-Y position of the selected curves for analysis .............................................................................. 19 Figure 5: Comparison of curve times for the three curves from session 1 (top) to 4 (bottom) (blue and red lines represent the mean curve times for experts and non-experts respectively). Each circle represents the curve time for one lap during a curve section from the respective participant. ................................. 22 Figure 6: Comparison of average lateral acceleration (g) for all the curves from session 1 to 4 (blue and red lines represent the mean curve times for experts and non-experts respectively). Each circle represents the mean lateral acceleration during a curve section from the respective participant.............................. 24 3 Figure 7: Session 4-Curve 1: Path Followed: Blue lines indicate the individual paths for the experts (right) and normal drivers (center) in session 4, whereas red line (for experts) and green line (for normal drivers) indicates the mean path of all the laps. Thin black lines indicate the lane boundaries of the track and the arrows indicate the direction of travel ......................................................................................... 27 Figure 8: Session 4-Curve 2: Path Followed: Blue lines indicate the individual paths for the experts (right) and normal drivers (center) in session 4, whereas red line (for experts) and green line (for normal drivers) indicates the mean path of all the laps. Thin black lines indicate the lane boundaries of the track and the arrows indicate the direction of travel. ........................................................................................ 27 Figure 9: Session 4-Curve 3: Path Followed: Blue lines indicate the individual paths for the experts (right) and normal drivers (center) in session 4, whereas red line (for experts) and green line (for normal drivers) indicates the mean path of all the laps. Thin black lines indicate the lane boundaries of the track and the arrows indicate the direction of travelthe track ........................................................................... 28 Figure 10: ISO Double Lane Change Maneuver.............................................................................................. 32 Figure 11: Adapted Double Lane Change Maneuver (Part A)......................................................................... 33 Figure 12: Oval Test Track (Part B)................................................................................................................. 33 Figure 13: X-Y Position of the selected curves for analysis ............................................................................ 34 Figure 14: Double lane change steering trace .................................................................................................. 37 Figure 15: Path followed during the Double Lane Change test (Black dots represent the cone position) ...... 39 Figure 16: Mean Steering Input for Experienced and Novice drivers ............................................................. 40 Figure 17: Repeatability in performance with respect to root mean square deviation in lateral acceleration and mean path (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) ...................................................................................................................... 42 Figure 18: Performance in terms of root mean square deviation from the mid path and the number of cones hit (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile).......................................................................................................................................... 42 Figure 19: Self-Assessment versus actual performance (number of cones hit) ............................................... 44 Figure 20: Comparison of Average steering rate and Average Steering Jerk between Novices and Experienced drivers (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile)..................................................................................................... 45 Figure 21: Maximum Steering Wheel Angle for the three steering inputs (See Figure 14). Top left represents the mean of the maximum steering angle in the first maneuver, top right represents the mean in the second maneuver and bottom represents the mean in the third maneuver (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) ............................ 46 Figure 22: Steering input comparison between Experienced and Novice driver............................................. 47 4 Figure 23: Mean lateral Acceleration at all speeds (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile).................................................................... 48 Figure 24: Deviation from inside of the curve: Top (left for novice and right for expert) is the path followed; Middle left is the best performance by experienced driver; Middle right is the best performance by novice driver; Bottom left is the mean performance by experienced driver; Middle right is the mean performance by novice driver .................................................................................................................. 54 LIST OF TABLES Table 1: Dependant Measures .......................................................................................................................... 20 Table 2: Number of Road Departures .............................................................................................................. 21 Table 3: Average curve-times for all session and all curves............................................................................ 21 Table 4: Average Lateral Acceleration for all sessions and all curves ............................................................ 23 Table 5: Steering metrics for all sessions and all curves.................................................................................. 25 Table 6: Braking point for all sessions and curves 2&3 (in curve 1 braking was not always observed)......... 26 Table 7: Distance traveled for all sessions and all curves................................................................................ 28 Table 8: Overview of the Test Conditions ....................................................................................................... 32 Table 9: Dependant measures used for analysis............................................................................................... 35 Table 10: Self Assessment score...................................................................................................................... 36 Table 11: Vehicle X-position when maximum steering input is given............................................................ 41 Table 12: Correlation coefficient of Self-Assessment versus number of cones hit ......................................... 44 Table 13: Steering Rate and Steering jerk (Steering Input C-E)...................................................................... 47 Table 14: Change in performance for 70 km/h double lane change ................................................................ 48 Table 15: Curve-Times (sec)............................................................................................................................ 51 Table 16: Mean Lateral Acceleration (g) ......................................................................................................... 51 Table 17: Steering Metrics ............................................................................................................................... 52 Table 18: Percentage Under-steer .................................................................................................................... 52 Table 19: Path Strategy (D1: Mean distance from inside of the curve; D2: Deviation from inside of the curve; D3: Deviation from the mean path................................................................................................ 53 5 ABSTRACT Safety is one of the major areas of concerns today in the field of automotive development. Different safety measures have and are being introduced in order to improve driver/passenger and pedestrian safety. Advanced driver assist systems (ADAS) are therefore becoming increasingly important in their role of reducing driver crash risk. A shortcoming of the ADAS systems is that the variability in drivers based on skill and experience is not taken into account and the system is often designed for average or worst case driver performance thereby compromising on the dynamic behavior of the vehicle. The present study focuses on understanding and quantifying the differences between expert, novice and normal (experienced) drivers in different driving situations. It is aimed to use this knowledge of driver differences in designing an adaptive ADAS by introducing the driver into the control loop. In the literature study it was found that higher steering activity in terms of steering rate, steering jerk, steering wheel angle and higher frequency steering inputs results in better performance and hence might be used to differentiate drivers based on skill level. In this study three experiments were performed to analyze differences between expert, normal (experienced) and novice drivers based on driver behavior and performance in dynamic steering task, one in a racing environment and two in a standardized driving environment. In the racing experiment previously obtained simulator data for 17 expert and normal drivers in a high-speed driving task on the Mallory Park test circuit was analyzed. The driving task required the participants to drive around the circuit to achieve the fastest lap times. Analysis showed that higher steering activity and differences in path strategy were the main reasons for lower lap-times shown by the expert race drivers compared to the normal drivers. Differences in steering behavior and path strategy can be attributed to differences in driving experience, vehicle dynamics knowledge and vehicle control skills. Steering metrics like average steering rate, steering jerk showed major and significant differences between the two groups. The second experiment compared experienced versus novice drivers in a double lane change (DLC) maneuver. This study was oriented around finding the differences between novice and normal experienced drivers while performing a double lane change maneuver at various speeds (from 70km/h to 105km/h). Data analysis showed that late initial steering input given by the novices compared to the experienced drivers was the main reason for their poor performance. This could be attributed to poor perception skill of the novices due to lack of knowledge and experience. Steering metrics like timing of steering input, average steering rate and average steering jerk showed statistically significant differences between the two groups. In the third experiment differences between novice and experienced drivers was investigated in a high-speed cornering task on an oval track. Experienced drivers showed better performance in terms of lower lap-times and higher average lateral acceleration as compared to the novices. Similar to the racing experiment, higher steering activity shown by experienced as compared to novices was the reason for better performance. Metrics like steering rate, steering jerk and steering reversal rate showed strong and significant differences between the two groups. Experienced drivers also showed better consistency in following their strategy. Summarizing, the experiments revealed that the three groups, experts, experienced and novices provide marked and strong differences in performance, steering behavior, path strategy and repeatability. Higher steering activity was found to be the main reason for better performance in the racing experiment and the high speed cornering experiment. Appropriate control inputs (steering angle) and accurate timing of steering input resulted in better performance in the DLC test. Steering metrics like steering jerk, steering rate, steering reversal rate and timing (position) of steering input showed significant differences between the groups and can be used to classify drivers based on skill level. Future research should be focused on using these results in a driver adaptive ADAS system and analyzing the differences in optimum performance. Furthermore different driving scenarios should also be researched to classify drivers based on skill level. Research should also focus on performing real vehicle experiments to validate the simulator experiment results. 6 CHAPTER 1: INTRODUCTION Through the years of automotive development safety has been one of the primary areas of concern. Inappropriate driver behavior and insufficient skill are considered the primary cause of road accidents. Brookhuis et al. (1987) stated that more than 85% of the total accidents could be directly attributed to the driver (59% driver error and 26% driver impairment due alcohol or drug consumption, fatigue etc.). Over the years there have been many regulatory and technological advances designed to help reduce the risk of accidents. Fildes and Lee (1994) reviewed several studies and reported a reduction of 8-40% in road accidents in Belgium, Finland, France, Germany, UK and South Africa due to reduced speed limits. They found that reduction in speed limits resulted in reduction of average speeds, which was responsible for reduction in the number of crashes and also severity of the crashes. Advanced driver assist systems (ADAS) are also becoming increasingly important in their role of reducing driver crash risk. Studies (Weiner and Curry, 1980; National Highway Traffic Safety Administration, 2009) have shown that these systems are beneficial in improving the safety of the driver and the passengers, but at the same time they can also make the drivers complacent and relaxed thereby increasing the probability of a dangerous situation. Moreover, these regulatory and technical measures are not driver specific and often the average driver performance is used as the benchmark. It can be argued that the present safety measures are not designed to address the variance in the skill level of drivers. Thus one of the challenges is to design ADAS systems to keep the vehicle safe and intuitive, while at the same time meeting the different demands and capabilities of different kinds of drivers. 1.1 DRIVING TASK Driving involves many tasks of varying levels of difficulty: complex tasks, such as driving on a winding hilly road, and easier tasks, like lane keeping on a highway. The driving task can be divided into the primary task, i.e. controlling the vehicle motion (vehicle heading, vehicle positioning and speed control), and secondary tasks, like guidance and navigation, and operating the in-vehicle information systems (Zhang et al., 2008). The driver is knowingly and unknowingly performing secondary tasks to fulfill the primary task, while keeping an eye on the road, the surroundings, and accordingly taking appropriate action when required. Driver situational awareness, his mental state, physical capability and responses affect the various driving tasks. Awareness and perception involve perceiving information using visual and other sensory cues. The driver mentally processes (organization and interpretation) the gathered information to evaluate the necessary responses. Physical reach, strength, flexibility and responsiveness then define how well the driver transforms the processed information into action. Concluding, the whole process of driving varies among individuals based on factors like sensory limitations, motor skills, learning, goals, personality, and previous experiences (Fuller, 2005; Malik, 2011). 7 1.2 SAFETY AND THE ROLE OF DRIVER ASSIST SYSTEMS As explained in the previous section, driving involves various concurrent tasks. Fuller (1984) mentioned that, compared to most other human tasks, driving has a higher probability of drivers encountering dangerous scenarios. Advanced driver assist systems (ADAS) have become increasingly important in their role of helping drivers to reduce crash risk. The National Highway Traffic Safety Administration (2009) completed a statistical analysis on accident data collected between 1995 and 2007 to estimate the effectiveness of Antilock Brake System for passenger cars and LTVs (light trucks and vans). It was found that ABS resulted in a reduction of 13% in fatal collisions with pedestrians and a reduction of 12% in fatal involvements with other vehicle on wet, snowy or icy roads. Although there was a significant 9% increase for ABS on session-off road crashes it was found that when ESC was combined with ABS there was a remarkable 30% reduction of fatal session-off crashes. Simpler ADAS systems (e.g. lane assist systems) help the driver to keep the vehicle under control by giving warning feedback to the driver through visual, audio or haptic cues, whereas more complicated systems like ESP are able to detect when a vehicle starts to skid and can then intervene to ensure that the driver/vehicle regains control. In these situations the ADAS system shares the driving task with the driver by giving corrective vehicle input to keep the vehicle under control. Although the ADAS systems have proved beneficial in improving the safety of driver and passengers, at the same time they can also make drivers complacent and willing to take risks, increasing the probability of dangerous situations. Weiner and Curry (1980) studied the promises and problems of automation and stated “one disturbing side effect of automation has appeared, i.e. a tendency to breed inactivity of complacency.” Another shortcoming of the ADAS systems is that they do not cater to the driver specific needs but often use average driver performance as the benchmark. Evans (2004) mentions that a driver chooses his own level of task difficulty and increased skill correlates to increased chosen task difficulty. Thus a high skill driver might choose a higher driving speed, with increased task difficulty, compared to a normal driver. ADAS systems however, aim at keeping the safety margins constant for all drivers. This may prevent high skill drivers from achieving the maximum possible performance. Early activation of the ADAS system affects the handling limits and hence the driving pleasure for a skilled driver. Continuous early activation (in a situation that an expert driver perceives as not dangerous) might cause nuisance for the driver, possibly causing the expert driver to switch off the ADAS. Original equipment manufacturers generally aim at keeping the vehicle on the safer side, thereby compromising dynamic behavior and handling limits. Ideally there should be different modes for different driver types, allowing the driver to select different levels of control authority. However, self-assessment problems could arise, as drivers might over or under estimate their skill level. 8 A solution is to gain knowledge of the differences in driver skill level and variations in driver control behavior and use this knowledge to design an ADAS system that would adapt to the driver requirements by introducing the driver into the control loop. The actual development of such a system is beyond the scope of this MSc thesis and could be the subject of a future assignment. 1.3 RESEARCH PROBLEM With high emphasis on driver and vehicle safety systems, knowledge of driver based differences in control and behavior can further improve the development of ADAS. The research hence focuses on quantifying differences between drivers with various skill and experience levels based on steering metrics in driving situations like lane change and cornering. Obtaining objective metrics capable of differentiating drivers can help in developing driver based adaptive safety systems. 1.3.1 RESEARCH STATEMENT This research is an extension of previous research done to classify drivers into groups of novice, experienced and experts. The research aims at objectively estimating the skill level of the drivers using predefined tasks of varying complexity. The study aims at focusing on driver behavior in conditions involving extreme steering and loss of control. The ultimate vision is to develop an ESC system tailored to the capability and preferences of individual drivers. The system is envisioned to judge the skill level of the driver based on his driving style (velocity, acceleration, braking, steering input, gear shifting etc). The control algorithm would then use the skill level as a variable to define the share of authority between the driver and the system. The actual development of such a system is beyond the scope of this MSc thesis. The main questions addressed in this research are: This research studies the correlation between driver performance and experience in different driving environment and maneuvers. The study was focused on finding main differences between experts, normal (experienced) and novice drivers. High speed driving experiment in a racing environment was analyzed to understand the differences between experts (race drivers) and normal drivers. Another experiment was performed to differentiate the novice and normal drivers in standardized driving situations like lane change and high speed cornering. The thesis focused on studying the steering behavior of the drivers. Past research has shown that there is a difference in steering behavior between experts, experienced and novice drivers. Moreover systems like ESC operate in extreme steering conditions when the vehicle is at the limit. Thus it is important to understand steering behavior of drivers with different skill level, especially in extreme driving conditions. Steering metrics like steering rate, steering jerk, steering reversal rate and steering input timing (position) were used to analyze the performance of the different group of drivers. 9 1.3.2 METHODOLOGY ADOPTED Overall driver competency and skill is based on driver awareness of the situation, perception, mental and physical capability and response. These factors are influenced by drivers’ biological characteristics (like information processing capacity and speed, reaction time, motor skills, physical reach, and flexibility and strength) and knowledge and skill (vehicle control skills, hazard detection and recognition skills and anticipatory and defensive driving skills) gained through experience. In the first half of the thesis (Chapter 3) data from a previously performed experiment was analyzed for driver-based differences in a racing environment. Performance of race drivers was compared with normal drivers in a simulator-based test. The second half of the thesis (Chapter 4) focused on studying the driver behavior during driving tasks like lane change and cornering. Performance of novice drivers was compared with experienced drivers in a simulator-based test. Moreover both the studies help understanding the behavioral differences over the whole spectrum of driving population divided into experts, experienced and novice. 1.4 THESIS OUTLINE This thesis comprises 6 chapters including introduction. The first chapter, as explained, has provided the research problem and thesis objectives. Chapter 2 reviews the previous research done in the field of driver skill evaluation. In Chapter 3, data from a previously performed experiment was analyzed for driver-based differences in a racing environment. Performance of race drivers was compared with normal drivers in a simulator-based test. Chapter 4 is divided into two parts, Part A and Part B. In Part A, performance of experienced and novice drivers was compared in a simulator-based double lane change maneuver. Part B studies the performance of experienced and novice drivers in a high-speed oval test track. This study helps in understanding differences between the two groups in extreme steering conditions. The adaptability behavior of the drivers was also studied by changing the coefficient of friction of the oval track in Part B. Chapter 5 concludes the thesis with the summary and discussion of the work done and the final results. Chapter 6 provides recommendations for future work in the field of driver skill evaluation. 10 CHAPTER 2: LITERATURE STUDY Differences in human behavior in executing various tasks have long been an area of interest among scientists and engineers. Research has been done to classify human behavior into expert, experienced and novice in various fields and tasks. In the context of driving behavior, a novice driver is one who is familiar with the task of driving but has limited driving experience. An experienced driver is one who has a certain level of expertise gained through driving experience, while expert drivers are categorized by high level of driving proficiency (e.g. race car drivers, instructors in driving school). Different driving tasks require different degrees of driving skill, situational awareness, and mental capacity to process the environmental information and furthermore require the physical ability to perform the response action. Malik (2011) defines driving competency as the ability to use the driving knowledge and skill for the successful and safe completion of the driving task i.e. proper and timely hazard perception followed by appropriate action. Fuller (2000) defined ranges of skills described as roadcraft, which he states, are the measure for a driver’s competence. These skills include vehicle control skills, hazard detection and recognition skills and anticipatory and defensive driving skills. Fuller (2005) also states that the initial driving performance is defined by personal biological characteristics like information processing capacity and speed, reaction time, motor skills, physical reach, and flexibility and strength. He further states that the competence increases as the driver gains knowledge and skill through training and experience. Both, biological characteristics and acquired knowledge and skill define total driver competency. Thus an expert driver is one with high competency. An inexperienced driver will have lower competency, which means that he/she might not be able to perceive a hazardous situation or maybe he/she is late in doing so. Fuller (2000) defined driving as a self-paced task i.e. the driver himself chooses speed, time headway etc. Evans (2004) states that a highly competent driver tries to maintain a higher speed, overtake in tight situations and tries to perform more secondary tasks. 2.1 PREVIOUS RESEARCH Hollopeter (2011) studied the difference in lane keeping performance during a simulator-based test of a lane departure avoidance scenario, between novice (16-18 years) and experienced (35-55 years) drivers. The results show that better and timely corrective actions taken by experienced drivers resulted in a lower average maximum lateral lane deviation (0.9 meters for the experienced drivers) as compared to the novices (2.2 meters). The author states that “Overall, the behavioral responses of the novice drivers were weaker than those of the experienced drivers”. The novices had lower steering activity in terms of initial steering response (21 degrees) and average steering jerk (14700 deg/s3) compared to the experienced drivers (30 degrees and 19600 deg/s3), which shows that possibly the novice drivers showed less urgency compared to the experienced drivers in reacting to the lane departure due to a poor perception of risk. 11 Zhang et al. (2008) used Discrete Fourier Transform (DFT) coefficients to differentiate between experts, experienced and novice drivers. The author states that on comparing the DFT of the steering signal from a simulator-based double lane change test, the experts show two major components, a slow one at 0.5 Hz and a faster one at about 1.1 Hz. On the other hand the novices did not show any high frequency component. This study suggests that experts and novices differ in steering input frequency, with experts operating at a higher frequency as compared to the novices. Matteo Rizzi (2005) studied the steering behavior of 44 drivers in double lane change (DLC) maneuvers on a slippery surface. The author uses Loss of Control scale (LoCS) and Lateral Adherence Ratio (LAR) in order to compare the performances of the drivers. LoCS varied from 0-3 based on number of cones hit and the steering correction used by the driver to keep the vehicle in control (0 being no steering correction and no cones hit and 3 being total loss of control). The author compensates for the difference in ice friction during the experiments using a calculated parameter called lateral adherence ratio (LAR). LAR is then calculated as the ratio of lateral acceleration during DLC test and average longitudinal deceleration during Combi-test (offset cone test). The study has contradictory results. The data suggests that the higher the value of LAR (i.e. a demanding maneuver) the higher should be the steering wheel speed to keep the vehicle in control. On the other hand the data also shows that the highest steering wheel speeds were achieved by drivers who lost control of the vehicle i.e. LoCS value of 3. The author explains that the reason could be that just a quick steering action is not enough to keep the vehicle in control but the driver needs to perform the steering action not too late. Katzourakis et al. (2011) studied differences in driver steering and throttle actions in a high-speed circular drive maneuver using an instrumented vehicle. Between the 3 drivers with racing experience (experts) and 3 drivers with no racing experience differences were found in steering control strategy and driver consistency. It was found that the expert drivers showed less deviation in vehicle states like velocity, yaw rate and slip angle. Experts furthermore had a lower deviation in steering wheel angle and higher deviation in throttle angle, which indicates that the drivers tried to control the vehicle using rear wheel slip (rear wheel drive test vehicle was used). Treffner et al. (2002) studied the difference in driving behavior of expert (driving instructors) and nonexpert, experienced drivers (more than 5 years of driving experience) during a variety of everyday maneuvers like cornering and high-speed swerve and recovery tasks using an instrumented vehicle. Expert drivers showed a definite apex while cornering, whereas the non-experts tried to follow the inside edge of the track. The experts followed a repeatable strategy of positioning far from the inside of the track while entering and exiting the corner. This is reflected in the standard deviation of the distance to the inside of the track, which was 0.31 m for experts and 0.57 m for the non-experts. 12 Erséus et al. performed a number of experiments to correlate driver experience to driver performance in various driving situation like curved cone track scenario to investigate drivers’ ability to steer the vehicle, i.e. path tracking skill (2007), line jump scenario to investigate driver steering behavior during sudden unexpected reference path movement (2008), avoidance maneuver in a double lane change (DLC) test to measure the driver skill based on different vehicle parameters (2009). His studies revealed different parameters that can be used to distinguish drivers based on their skill level. Curved cone track scenario showed that cornering strategy is a good measure for analyzing the drivers’ decision making, while the magnitude of the standard deviation from the average path is useful for analyzing the repeatability of the drivers, and thereby provides an indication of the precision with which the driver controls the vehicle. In the line jump scenario it was seen that high skilled drivers had lower overshoot and a short rise time compared to the low skill drivers. The small overshoot (accurate and precise inputs) along with a small rise time (quick and timely response) shown by the high skill drivers resulted in a stable situation after the line jump, as compared to the low skill drivers High skilled drivers also showed a lower standard deviation in lateral velocity and yaw angle. In the DLC-Test, it was found that only one low skill driver achieved a maximum speed of 100 km/h (50km/h: 1 driver, 60km/h: 2 driver, 70km/h: 4 driver, 80km/h: 3 driver, and 90km/h: 1 driver) whereas all the high skill drivers achieved a maximum speed of 90 km/h or above. Lane positioning (y-position) was found to be a good differentiating parameter at speeds of 50, 60 and 70 km/h. The author Standard deviation of the steering wheel rate and the standard deviation of the angular acceleration showed differences between the high skill and low skill drivers. Kinnear et al. (2012) studied how different drivers anticipate risk on the road using skin conductance responses (SCR). The test subjects were categorized into learner (mean of 150 km of driving in the past 12 months), inexperienced (mean of 5000 km of driving in the past 12 months), and experienced group (mean of 10000 km of driving in the past 12 months). The groups were shown clips of hazardous driving situations and they were compared based on their SCR ratings. An SCR value of 0.05 µS is counted as a response and the final SCR score is calculated using the formula: Score(%) = no. of clips with an SCR "100 Total no. of cilps SCR value above 0.05 µS is counted as one response irrespective of the amplitude of the value. A score of 0% means that the participant did not respond to any stimuli. This can be because he was not gathering ! information from the right areas (visual perception) or he was incapable to judging the severity of the situation i.e. low risk perception level (due to lack of knowledge and experience). A score of 100% indicates a highly alert and aware participant with a high risk perception level. The group responses were divided into anticipatory and event responses. It was found that the learner group showed the minimum response level 13 with an anticipatory SCR score of around 24% and the event SCR score of around 42%. The inexperienced group had an anticipatory SCR score of around 33% and event SCR score of around 61%. The experienced group showed the highest level of response with an anticipatory SCR score of around 66% and event SCR score of around 80%. Thus the study clearly shows that an experienced driver has a better anticipation of the upcoming hazard than inexperienced and learner drivers. The study however does not investigate whether this risk anticipation of the hazard actually results in a safer behavioral response when driving. 2.2 INTELLIGENT ADAS SYSTEMS Some recent studies have been focusing on including the driver variances while designing the driver support systems. Miller et al., (2003) mention in their study that, “One of the challenges here is that there is a large variance among different drivers in their driving behavior, and thus the driver support system should be able to adapt to the peculiarity of each driver.” They design a Driver Advocate system (DA), which is classified into driver monitoring, user modeling and interacting with the driver. Driver monitoring involved studying the driver response and behavior (steering, braking, acceleration, eye movement etc.). This helps the system understand where the driver directs his visual attention and to predict whether the driver understands the situation and makes assumptions about what driver intentions are. Driver modeling is used to update the system for different drivers. The DA system communicates with the driver providing warning information in time for the driver to react. The author narrates a situation in which the driver has looked at a bicycle 10 meters ahead. The driver can react to the situation by either reducing speed or by changing lanes. If neither action is taken within a stipulated amount of time (calculated by the DA based on time to collision), then the DA infers that either the driver is at fault or there is an error in the user model of the DA (i.e. there is no real problem perhaps because the bicycle is in the bicycle lane). The current mentioned DA assumes that all the errors are driver errors and thus provide an audio, visual or haptic feedback. Author states that the system does not consider that there are some drivers that tend to correct as late as possible. The model is just used to validate how well an upcoming situation can be predicted so that the driver can be alerted of the situation. Cacciabue et al. (2009) used a similar approach to design an adaptive and intelligent system that is capable of predicting between-driver variability. The study proposes an “active” driver model that fulfils the following requirements: 1) The model is capable of completing all aspects of driving task (i.e. control, guidance and navigation), 2) The model is capable of representing both behavioral and cognitive aspects of decision making and 3) The model capture driver response and decision making, given the driver type, driver state and situation. This driver model simulation is used along with environment and vehicle model simulation in a Driver-Vehicle-Environment (DVE) interaction model. The model was then simulated for two tasks 1) steering task and 2) dynamic speed and acceleration. In the speed task the simulation was session on an urban road at a speed of 90 km/h. The variables used were steering angle, lateral distance from 14 centre-line and lateral acceleration. For the speed test the simulation varied the speed along the whole road based on model parameters, road (geometry, condition, etc.) and traffic conditions. In order to validate the simulation results it was compared with the results from a similar test performed in a driving simulator at the Swedish Transport Research Institute (VTI). On comparing the steering wheel angle data, a maximum deviation of 30% was found, which was well within the deviation that the simulator test showed. The speed validation was done by simulating three conditions a) risky and aggressive driving, b) very cautious driving and c) average driver behavior. Varying the parameters like experience, attitude, task demand etc simulates these behaviors in the driver model. The driving simulator test data showed a mean speed value, which lies between the case (a) and case (c). A non-linear relationship was found between the results from the three cases. The comparison validates the results of the simulation and shows capability of predicting between driver variability. The next step is to create a driver model that can identify the driver characteristics from real-time observed data. 2.3 SUMMARY Thus it can be summarized that compared to experienced and novice drivers, expert drivers show higher frequency steering inputs and higher steering activity (in terms of steering rate, steering jerk and steering wheel angle), which results in better performance. The exact reason for this behavior is not fully understood and it can be because of high bandwidth control shown by experts. Experts and experienced drivers also show a consistent and precise performance while following their strategy, which is reflected in smaller deviations shown by experts during lane keeping performance and cornering. One of the weak points of the studies reviewed was that they merely categorize drivers into experts, experienced and novices. Most of these studies compare the various groups to find significant differences in performance parameters and are not capable of predicting whether the driver is an expert or novice by just analyzing his performance. Moreover the studies reviewed find differences between experts (generally race drivers) and experienced or experts and novices. Not much research has been focused on finding differences between novice and experienced drivers. The review also revealed some areas of driver behavior where not much research has been done. None of the studies tried to correlate the motor skills and vehicle dynamics knowledge of the driver to performance in the real tests. Drivers proficient in these areas are expected to produce better performance. Another area of driver behavior that has not been studied is the performance in loss of control situations. Analysis of the driver actions that lead to loss of control situations can help us recognize the areas of driver behavior responsible for such situations (e.g. over speeding, imprecise vehicle control, lack of vehicle dynamics knowledge shown through improper combination of control inputs etc.). Another area of interest that has not been researched in detail is driver adaptability to change in vehicle and environmental conditions. 15 Effect of driver variances in designing driver support systems has been recognized in some studies but not much work has been done to scale/rate the expertise of individual drivers. Having such information can help design driver assist systems adapted to the needs of individual drivers. It is a complicated area of research due to the vast variance in driver skills and driving scenarios. One of the challenges is to choose what scenario would provide the best insights to identify a driver’s skill. Another challenge is to move from qualitative to quantitative analysis of completion of the task i.e. to quantify how well a task has been performed. The ultimate challenge is to correlate the task performance to the driver skill scale/rating. 16 CHAPTER 3: RACE VERSUS EXPERIENCED DRIVERS ABSTRACT In this study differences between race-car drivers and normal drivers have been investigated in a high-speed driving task. The study analyses previously recorded simulator data for 17 drivers on the Mallory Park test circuit. The driving task required the participants to drive around the circuit to achieve the fastest lap times. Analysis showed that higher steering activity and differences in path strategy were the main reasons for lower lap-times shown by the expert race drivers compared to the normal drivers. Steering metrics like average steering rate, steering jerk showed higher values for the expert group and distance traveled around the corner showed a different path strategy adopted by the experts. Both groups showed improvement in performance based on lap-times across the different sessions. Thus the study shows that experts and normal drivers have different steering behavior and path strategy, which can be attribute to differences in driving experience, vehicle dynamics knowledge and vehicle control skills. 3.1 INTRODUCTION Driver perception, knowledge and awareness, and vehicle control skills affect the driving task competence. Fuller (2005) stated that the total driving competence is based on initial driving skill and the knowledge and vehicle control skills gained through experience. Malik (2011) defined driver competency as the ability to use the driving knowledge and skill for the successful and safe completion of the driving task i.e. proper and timely perception followed by appropriate action. Previous research has shown differences in control strategy for expert drivers compared to non-experts (Katzourakis et al., 2011) on a circular drive task. Further differences in steering control in a simulated environment were found by Zhang (2008), who showed differences in the frequency spectrum between expert and novice drivers during different lane change maneuvers. The difference in competency level between experts and normal drivers leads to differences in behavior and performance. The present chapter focuses on analyzing the differences between race drivers (experts) and normal drivers in a high-speed driving task. The task required the participants to select the optimal speed and race line, and provide accurate and consistent control inputs while going around the corners in order to achieve minimal lap-time. The study focuses on the steering behavior and the path strategy and consistency of both groups of drivers using a previously obtained dataset. 3.2 METHODS 3.2.1 APPARATUS The experiment was conducted in a race-car simulator based on the chassis of a Formula Renault 2.0 racing car used for training purposes (Sim-Delft, 2013). The steering wheel, brake and throttle pedal were used from the original car and a direct drive motor provided force feedback. The throttle and brake pedal feedback 17 was passive and calibrated to resemble a realistic formula car. The visual system consisted of three 52-inch LCD screens and provided a 130 degree horizontal and 27 degrees vertical field of view. The simulator was equipped with a steering wheel mounted dashboard showing speed, engine rpm, lap and lap sector times. The virtual environment, vehicle dynamics and force feedback were simulated by rFactor software (v1.255). The rTrainer vehicle model, a rear wheel driven formula style racecar (115 bhp, 573 kg), was used (Figure 1). All driving aids were disabled and gear shifting was automated. All driving simulator data was recorded and stored at 100Hz. Figure 1: RTrainer car Figure 2: Mallory Park Test Circuit 3.2.2 EXPERIMENT INSTRUCTIONS Participants were instructed to drive the fastest lap-time possible on an unfamiliar racetrack. The Mallory Park circuit was chosen for the experiment. Figure 2 above shows the outline of the circuit, consisting of: 1) a long right hand corner known as the Gerard’s bend, which turns through nearly 200 degrees, 2) a 180 degrees hairpin corner know as the Shaw’s Hairpin, and 3) a combination of two fast corners. The participants drove four sessions of 10 minutes and between sessions had a five-minute break. Participants received instructions prior to the start of the experiment about the application of the throttle and gas pedal and explanations regarding the information available on the steering wheel mounted dashboard. Participants were required to steer, accelerate and brake (gear shifting was automatic). 3.2.3 PARTICIPANTS Seventeen participants (all male), aged from 17 to 26 years (mean = 20.8, SD = 2.0) participated in the study. The non-expert group consisted of 10 participants, students at the Delft University of Technology having no experience in racing with an average age of 21.6 years (SD = 1.8). The expert group had 7 participants, professional racing drivers from various (international) racing classes (e.g. Formula 3, GP2 and Porsche Supercup) with an average age of 19.9 years (SD = 2.0). 18 3.2.4 DEPENDANT MEASURES To analyze the performance of the participants three curves were selected on the test circuit: 1) Long right hand curve, which turns through nearly 200 degrees 2) Combination of two fast curves 3) 180 degrees hairpin curve The figure below shows the selected curve and each curve was analyzed separately. Data analysis and the results are shown below. Figure 3: Track breakdown into different curves for analysis Each session was analyzed separately per curve. The data for the three mentioned curves was selected for every lap and analyzed. The curve entry and exit point was determined by the X-Y position of the vehicle, ensuring that it was situated before the point where drivers start giving any vehicle control inputs (steering, brake, or throttle) relating to the entry of the corner. Similarly, in determining the exit point it was verified that the vehicle was in straight-line steady state condition (Figure 4) Figure 4: X-Y position of the selected curves for analysis 19 The steering data was filtered using a low pass Butterworth filter (2nd order, 3Hz). All the other data was filtered using a Butterworth 2nd order 10Hz low pass filter. The first part of the analysis was to remove road departures from the data. The cases in which all the four wheels left the track were considered as “road departure”. This was done separately for all the three curves and laps with road departures were excluded from the performance analysis and were studied separately to understand the crash behavior of the two groups but has not been included in this thesis (see appendix 3). The performance of the drivers was analyzed using different dependant measures discussed in Table 1 below. As the data was found to be non-normal (using q-q plots), differences between the experts and the non-experts were assessed using Wilcoxon-Mann-Whitney test, which is better at dealing with non-normal distributions than the t-test. Table 1: Dependant Measures Metric Average Steering Rate Average Steering Jerk Mean Lateral Acceleration Curve Time Distance Covered Braking Point Description Average steering wheel rate from the initial steering input (Absolute Steering angle>3) to the curve exit point Average rate of change of the steering wheel acceleration (jerk) from the initial steering input (Absolute Steering angle>3) to the curve exit point Mean of the lateral acceleration from the entry to the exit point of the curve Total time taken from the entry to the exit point of the curve Total distance travelled from the entry to the exit point of the curve Total distance between the points when the driver first brakes after the curve entry point to a predefined X-Y position on the curve (same for all the drivers). Only Curve 2 and Curve 3 are studied for this measure, as these are low speed corners, which require the drivers to brake. Curve 1 was excluded from the analysis because not all drivers brake before or during Curve 1 but instead just released the throttle to reduce the speed. Units deg/sec (deg/s)3 g seconds meters meters 3.3 RESULTS 3.3.1. ROAD DEPARTURES Experts showed a higher percentage of road departures compared to normal drivers in Curve2, session 2-4. A significant improvement in number of road departures over sessions can be seen for the normal drivers in all curves (p=0.031), while the experts remain at a constant level of road departures (p=0.343). Table 2 below shows the number of laps with road departures for every session and both groups. 20 Table 2: Number of Road Departures Session Curve 1 Experts Curve 2 P value 0.108 Experts Curve 3 1 8 NonExperts 21 P value 0.718 Experts 18 NonExperts 18 10 NonExperts 25 P value 0.752 2 8 10 0.709 29 17 0.008 9 14 0.538 3 12 10 0.302 25 7 0.008 7 13 0.679 4 11 13 0.946 24 8 <0.001 7 11 0.923 Mean 9.25 (1.5) 13.50 (5.20) 0.171 24.00 (4.55) 12.50 (5.80) 0.057 8.25 (1.50) 15.75 (6.29) 0.029 Total Crashes* 72/125 (58%) 70/147 (48%) 60/148 (41%) 65/153 (43%) 66.75/143.25 (47%) * Total number of road departures for experts and non-experts combined / Total number of laps 3.3.2 CURVE-TIMES Experts show lower lap-times and curve times compared to the novices. Experts are faster than novices on average across all the sessions by a margin of 0.8-1.5 seconds in curve 1, 0.4-0.9 seconds in curve 2 and 1.02.3 seconds in curve 3. Both the groups show near significant improvement in lap times from session 1 to 4. The two groups perform significantly differently in terms of curve times from Session-2 to Session-4, as can be seen from the p-values indicated in the table below. Table 3: Average curve-times for all session and all curves Session 1 2 3 4 Session 1 to 4 difference Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts (p) Non-Experts (p) Total lap-time 56.58 (1.78) 60.99 (5.90) <0.001 55.71 (1.41) 58.85 (2.94) <0.001 55.13 (0.37) 59.80 (4.68) <0.001 55.25 (1.11) 59.36 (3.91) <0.001 0.072 0.066 Curve 1 18.02 (0.53) 18.61 (.46) 0.082 17.59 (0.42) 18.89 (1.12) 0.01 17.40 (0.32) 18.90 (1.08) <0.001 17.47 (0.21) 18.54 (1.02) 0.001 0.063 0.063 Curve 2 9.33 (0.30) 9.64 (0.26) 0.082 9.14 (0.35) 9.93 (0.97) 0.03 8.95 (0.10) 9.88 (0.84) <0.001 8.95 (0.09) 9.80 (0.69) <0.001 0.031 0.063 Curve 3 12.65 (0.36) 14.58 (2.46) 0.052 12.17 (0.19) 14.48 (3.64) 0.001 12.14 (0.19) 13.12 (0.44) 0.001 12.17 (0.20) 13.34 (1.18) <0.001 0.031 0.063 The overall difference in curve-times for all the three curves between experts and normal drivers is around 2.2 -4.6 seconds whereas the difference in lap-times is between 3.2-4.7 seconds. This shows that the experts maintained a higher speed in the straight road segments also, which maybe because of the higher corner exit speeds of the experts. Table 3 above does not indicate the improvement in the average curve times for curve 1 from session 1 to 4, which can be seen in Figure 5 below. 21 Figure 5: Comparison of curve times for the three curves from session 1 (top) to 4 (bottom) (blue and red lines represent the mean curve times for experts and non-experts respectively). Each circle represents the curve time for one lap during a curve section from the respective participant. 22 Similar differences are shown in curve 2 and curve 3 also with experts having lower curve times as compared to normal drivers. Both the groups show improvement across the sessions. Experts show significant improvement in curve-times in curve 2 and curve 3 between session 1 and session 4. Experts show more consistency in the results as can be seen from Figure 5 above. 3.3.3 LATERAL ACCELERATION Experts maintain higher levels of lateral acceleration as they went around the curves during sessions 1 to session 4 compared to the normal drivers. In curve 1 the expert drivers maintain 0.06-0.12 g higher lateral acceleration, in curve 2 the difference is 0.07-0.14 g and in curve 3 there is 0.07-0.09 g difference as compared to the normal drivers. No significant improvement was seen in the performance of the two groups in terms of lateral acceleration from session 1 to session 4 for the novices whereas the experts show significant increase in lateral acceleration for curve 1 from session 1 to session 4. Table 4: Average Lateral Acceleration for all sessions and all curves Session 1 2 3 4 Session 1 to 4 difference Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts Curve 1 0.77 (0.05) 0.71 (0.04) 0.126 0.80 (0.03) 0.69 (0.06) 0.01 0.82 (0.03) 0.70 (0.06) <0.001 0.82 (0.02) 0.72 (0.07) 0.002 0.031 0.313 Curve 2 0.71 (0.07) 0.64 (0.06) 0.125 0.78 (0.04) 0.65 (0.08) 0.003 0.78 (0.03) 0.64 (0.09) <0.001 0.78 (0.03) 0.64 (0.08) 0.004 0.156 0.313 Curve 3 0.47 (0.04) 0.40 (0.04) 0.03 0.49 (0.03) 0.40 (0.05) 0.004 0.497 (0.02) 0.42 (0.03) <0.001 0.49 (0.02) 0.42 (0.03) 0.002 0.219 0.188 Novices have lower values of acceleration probably because their primary goal is to successfully negotiate the curve without any road departures, whereas experts try to keep the vehicle at the limit to achieve the best performance. Thus experts also have higher road departures as compared to normal drivers (Table 2). 23 Figure 6: Comparison of average lateral acceleration (g) for all the curves from session 1 to 4 (blue and red lines represent the mean curve times for experts and non-experts respectively). Each circle represents the mean lateral acceleration during a curve section from the respective participant. 24 3.3.4 STEERING PERFORMANCE Experts show higher steering activity compared to normal drivers based on steering wheel rate and steering jerk. Steering jerk values are different for the two groups but only curve 1 (session1 and session 2) and curve 2 (session 1, session 3 and session 4) show significant differences (p < 0.05). Steering jerk values shown by the experts were approximately 1.5-2 times higher than the normal drivers for curve 1 and 2. It can also be seen that there is a reduction in the average steering jerk values for the expert group from session 1 to 4 for all the curves but only curve 3 shows significant difference (p<0.05). Normal drivers also show reduction in steering jerk values but it was not statistically significant. Data also shows high standard deviation among the experts. Normal drivers have lower values but also show smaller deviations in the group, see Table 5 for an overview of the data. Average steering rate is also higher for the expert group compared to the normal drivers but it is not significantly different. There is a reduction in average steering rate values for the normal drivers from session 1 to 4 but it is not statistically significant. Experts also show reduction but only curve 2 shows significant difference (p<0.05). Table 5: Steering metrics for all sessions and all curves. Session Curve 1 Steer Jerk 4791 (2139) 2378 (608) 0.017 Steer Rate 22.10 (10.57) 14.94 (5.63) 0.247 Steer Jerk 5216 (2876) 2657 (687) 0.017 Steer Rate 28.47 (8.44) 31.24 (6.21) 0.429 Steer Jerk 5904 (1342) 5328 (759) 0.247 18.33 (9.78) 10.85 (4.03) 0.09 3923 (2542) 1943 (843) 0.03 16.5 (6.01) 13.8 (3.41) 0.44 3783 (1562) 2302 (747) 0.07 24.9 (6.96) 23.5 (6.56) 0.9 5255 (1339) 4229 (1557) 0.3 14.87 (7.62) 10.48 (4.02) 0.27 3092 (2025) 1805 (708) 0.11 15.7 (5.85) 11.7 (2.27) 0.08 3629 (1285) 2038 (702) 0.01 21.45 (6.53) 20.89 (6.8) 0.96 4660 (1336) 3860 (999) 0.22 NonExperts p-value 14.44 (4.54) 10.05 (3.15) 0.07 2912 (1306) 1733 (631) 0.07 14.7 (4.5) 12.1 (3.67) 0.07 3188 (815) 2025 (1032) 0.023 21.82 (6.2) 20.32 (5.42) 0.92 4407 (1469) 3467 (1187) 0.25 Experts 0.219 0.094 0.031 0.063 0.156 0.031 NonExperts 0.188 0.313 0.313 0.188 0.188 0.188 NonExperts p-value Experts 2 NonExperts p-value Experts 3 NonExperts p-value Experts 4 Session 1 to 4 difference Curve 3 Steer Rate 21.82 (11.31) 13.28 (4.79) 0.330 Experts 1 Curve 2 25 3.3.5 PATH STRATEGY BRAKING POINT Lower values of standard deviation, especially in Curve 2, show that experts find a suitable braking point and follow it consistently. As can be seen from the Table 6, the experts brake approximately 94 meters before the curve 2 entry point during all the 4 sessions. On the other hand the normal drivers brake later and show variation in the braking point through the different sessions. Normal drivers also have a higher standard deviation within a session compared to the experts. The difference in braking point in Curve 3 is later for the experts, and reaches significance in the last session. No significant difference was seen in experts or the normal drivers from session 1 to 4 (p>0.05). Table 6: Braking point for all sessions and curves 2&3 (in curve 1 braking was not always observed) Session 1 2 3 4 Session 1 to 4 difference Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts Curve 2 93.82 (0.76) 93.75 (0.31) 0.126 94.1 (0.16) 92.9 (2.25) 0.03 94.08 (0.12) 91.8 (5.64) 0.003 94.07 (0.14) 93.55 (0.5) 0.01 0.999 0.125 Curve 3 355.29 (1.11) 356.25 (0.63) 0.052 355.07 (1.07) 355.88 (1.02) 0.11 355.2 (0.82) 356 (0.9) 0.07 355.13(0.6) 356.2 (0.9) 0.01 0.844 0.313 PATH STRATEGY In figures 7 to 9 the path taken by experts and normal drivers is shown for all curves in session 4. Differences in path strategy between the two groups can be seen from the graphs. In curve 1, as can be seen from Figure 7, normal drivers try to keep the vehicle close to the inside of the track at all times. Experts on the other hand first go towards the outside of the curve and then give a sharp steering input to exit from the curve. As a result, they maintain a larger distance from the inside of the curve as compared to the normal drivers. 26 Figure 7: Session 4-Curve 1: Path Followed: Blue lines indicate the individual paths for the experts (right) and normal drivers (center) in session 4, whereas red line (for experts) and green line (for normal drivers) indicates the mean path of all the laps. Thin black lines indicate the lane boundaries of the track and the arrows indicate the direction of travel In curve 2, as seen from Figure 8, normal drivers give a higher initial steering input to keep the vehicle closer to the inside of the track. Experts, on the other hand, have smaller steering inputs and keep the vehicle closer to the outside of the track while exiting the corner, compared to normal drivers. Figure 8: Session 4-Curve 2: Path Followed: Blue lines indicate the individual paths for the experts (right) and normal drivers (center) in session 4, whereas red line (for experts) and green line (for normal drivers) indicates the mean path of all the laps. Thin black lines indicate the lane boundaries of the track and the arrows indicate the direction of travel. In Curve 3 as seen from Figure 9, the non-experts try to remain close to the inside of the curve while entering and try to maintain a constant distance from the inside of the corner. The experts on the other hand, drive away from the inside of the curve while entering and remain close to the inside of the curve while exiting. 27 Figure 9: Session 4-Curve 3: Path Followed: Blue lines indicate the individual paths for the experts (right) and normal drivers (center) in session 4, whereas red line (for experts) and green line (for normal drivers) indicates the mean path of all the laps. Thin black lines indicate the lane boundaries of the track and the arrows indicate the direction of travelthe track Difference in path strategy is also represented in the distance travelled by the two groups while negotiating the corner (Table 7). As can be seen from Table 7, experts take a longer path in curve 1 (significantly different for Session 2 and Session 4 with p<0.05) which is because they maintain a higher distance from the inside of the curve (see Figure 7) as compared to normal drivers. Similar pattern is seen in curve 2 (significantly different for Session 3 and Session 4 with p<0.05) where the experts keep the vehicle towards the outside of the curve while exiting (See figure 8). Curve 3 does not show any significant difference in terms of distance travelled. There is no significant difference for distance travelled from session 1 to 4 for both the groups. Table 7: Distance traveled for all sessions and all curves. Session 1 2 3 4 Session 1 to 4 difference Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts p-value Experts Non-Experts Curve 1 723.81 (1.91) 722.54 (0.92) 0.429 726 (1.45) 723.2 (2.17) 0.02 725.6 (2.06) 723.4 (2.08) 0.07 727.4 (2.17) 723.7 (2.14) 0.01 0.063 0.125 Curve 2 359.97 (4.29) 355.65 (2.07) 0.052 361.2 (3.8) 357.01 (3.56) 0.07 360.12 (3.06) 356.7 (3.28) 0.050 360.14 (2.16) 356.2 (3.4) 0.01 0.844 .063 Curve 3 306.02 (4.40) 306.53 (3.96) 0.999 304.05 (8.18) 303.62 (6.96 0.9 304.41 (6.04) 303.92 (5.8) 0.73 305.07 (5.4) 307.03 (6.96) 0.68 0.188 0.999 28 3.4 DISCUSSION In this chapter differences between expert race-car drivers and normal drivers were investigated. As seen from the data presented in this chapter, experts show better performance in terms of lap-times and higher tire utilization as compared to the normal drivers. Higher lateral acceleration shown by the expert drivers means that they had a higher speed throughout and exiting the corner. This enabled them to maintain a higher straight-line speed, resulting in faster lap-times achieved by the experts as compared to the normal drivers. Normal drivers had more road departures in the first session, which could be because they were not familiar with the simulator and took some time to adjust to driving in the simulator environment, and getting accustomed to the racing task and the race-track. In sessions 2 to 4 experts had higher percentage of road departures as compared to the novices, which is possibly because experts took more risk trying to keep the vehicle on the limit to achieve best performance as can be seen from the lower lap-times and higher tire utilization values, compared to the normal drivers. Probably, the primary goal of the normal drivers was to successfully negotiate the curve without any road departures, sacrificing some curve time performance. The results show differences in steering behavior between the two groups. Expert drivers show higher steering activity in terms of steering rate and steering jerk. These results are consistent with previous research as discussed in chapter 2 which showed higher steering activity among expert drivers in terms of steering wheel angle, average steering jerk and frequency of steering inputs (Hollopeter, 2011; Zhang et al., 2009). Driving is a combination of open and closed loop processes. Precise timing and accurate control inputs can enable the driver to negotiate the race task in a largely open loop state. While entering a corner a driver anticipates the speed and steering angle that should be used, representing the feed-forward part of the control loop. But deviation from the desired path (vehicle positioning on the track) due to imprecise control inputs, lag in the system or changing vehicle and environment conditions, requires for additional control inputs to correct for the deviations from the desired path. This represents the feedback control of the drivers. Possibly best performance can be achieved with a combination of feed-forward and feedback control. Therefore, the higher steering activity shown by the experts might be attributed to higher feed-forward and feedback gain as compared to the normal drivers. Lower steering activity shown by the normal drivers might be correlated to lower feed-forward gain, which resulted in poor vehicle positioning while entering the corner, and lower feedback gain, which resulted in insufficient correction in vehicle path while taking the corner. Thus higher steering activity shown by the experts can possibly be attributed to optimizing the desired path. Another possible explanation for higher steering activity especially in the racing task can be that the experts are not only optimizing the path followed but are also trying to keep the vehicle at the traction limit and hence providing continuous steering corrections to stabilize the vehicle. Overall this is evidence that experts have a better internal vehicle model developed, which enables them to understand what the current situation demands, what should be the control inputs, and also how the vehicle will respond to the 29 given control inputs. The promptness in steering action and the ability to provide faster inputs could be the result of practice. Normal drivers on the other hand might not be capable of giving such fast inputs, lacking motor control skills or maybe they do not dare to give faster inputs as they do not know how the vehicle might respond which is evidence of inferior vehicle dynamics and response knowledge. This is consistent with the definition of competence (Fuller, 2005), which states that competency is a combination of initial personal biological characteristics and knowledge and skill gained through training and experience. We also see a difference in strategy between the two groups in terms of braking point and the path chosen to negotiate the corner. The normal drivers show inconsistency in the braking point. This can be evidence that the novices are inaccurate in perceiving the curvature of the corner and hence are unable to judge the correct timing and magnitude of control action. While cornering, normal drivers try to maintain a constant distance from the inside of the corner whereas the experts tried to follow the racing lane, keeping towards the outside of the corner while entering and exiting and going close to the inside of the corner in the mid-section. The different path strategies of the two groups is similar to the one found by Treffner et al. (2002). In summary, it can be concluded that experts who had greater experience in the racing environment performed better than the normal drivers in terms of lower lap-times. Higher steering activity, different braking and path following strategy and consistency in following the chosen strategy significantly differentiated the two groups. The data also showed that lap-time, steering jerk and distance travelled metrics could be used to differentiate between expert and normal drivers. Steering jerk metric showed the largest difference between the two groups with experts approximately 1.5-2 times higher than the normal drivers for curve 1 and 2. 30 CHAPTER 4: NOVICE VERSUS EXPERIENCED DRIVERS ABSTRACT This study was oriented around finding the differences between novice and normal experienced drivers while performing a double lane change maneuver and a high-speed cornering task. The study aimed at finding parameters capable of differentiating the two groups with special emphasis on steering behavior. Part A of the test procedure required the participants to complete a double lane change at various speeds (from 70km/h to 105km/h). Data analysis showed that late initial steering input given by the novices compared to the experienced drivers was the main reason for their poor performance. Steering metrics like timing of steering input, average steering rate and average steering jerk showed statistically significant differences between the two groups. Part B of the experiment required the participants to drive around a flat oval track to achieve the fastest lap times. Analysis showed that higher steering activity and differences in path strategy were the main reasons for lower lap-times shown by the experienced drivers compared to the novice drivers. Steering metrics like average steering rate, steering jerk showed higher values for the experienced group. 4.1 INTRODUCTION In the previous chapter we studied differences between expert and normal (experienced) drivers in a highspeed driving task on a racing circuit, driving on the traction limit. This chapter focuses on analyzing differences between novice and experienced drivers in standardized driving situations such as lane change and high speed cornering which are relevant to normal driving. Thus in this chapter we try to study the behavior of the young inexperienced drivers (novices) in normal driving situations. Details of the experiment have been explained below. 4.2 METHODS 4.2.1 X-CAR SIMULATOR The driving simulator used for the experiment is based on a dSPACE real-time (RT) computer. It executes a commercial RT vehicle-dynamics model (VDM), developed on an open MATLAB/Simulink block, from the dSPACE Automotive Simulation Model (ASM) package. The simulator consists of a dSPACE computer with two DS1005 boards in master–slave topology. The dSPACE interfaces with the environment via A/D (DS2002) and D/A (DS2102) boards. One desktop PC is used as a user station providing the interface to the simulator. It is also used to develop the VDM and control algorithms. The dSPACE simulator transmits the animation data over an Ethernet connection to three desktop PCs handling the graphics. Finally, three LCD monitors compose a horizontal viewing angle of 135 deg were used to display the visuals. A high-response ac brushless servomotor (Ultract II, type 708303) is used for providing 31 steering force feedback (Katzourakis et al., 2011). The brake and throttle pedal originated from a real vehicle providing realistic passive feedback. All driving simulator data was recorded and stored at 100Hz. 4.2.2 EXPERIMENT INSTRUCTIONS, PART A: DOUBLE LANE CHANGE The experiment is divided into two parts, Part A and Part B. In Part A of the experiment, participants were instructed to drive a double lane change maneuver adopted from ISO 3888-1:1999-Part 1: Double lane change, which is shown in Figure 10 &11 below. The lane change track has three lanes depicted by the cones, lane 1 is 15 m, lane 2 is 25 m and lane 3 is 15 m in length. Participants were instructed to keep the vehicle within the lane boundaries (depicted by the cones) during the complete maneuver. As compared to the ISO procedure the lane width was increased to 3 m, in order to make experiment completion feasible also for drivers without specific training. Before starting the testing trials a practice session of 5 minutes was given such that the participants were accustomed to the simulator environment and the test condition. The speed control for the experiment was automatic and thus the participant could only control the vehicle using the steering wheel. The experiment was started from a speed of 70km/h and 5 trials were given at 8 different speeds. After each trial the vehicle was automatically reset to the starting position and the next trial was started. After the 5 trials of each speed a 1-minute break was given during which the participant remained seated in the simulator. Speed increments were done in steps of 5km/h to 10km/h up to a speed of 105 km/h. In the end the experiment was repeated at 70 km/h to retest the first speed condition. Table 8 gives an overview of the test conditions. It was hypothesized that performance would degrade with speed and that this degradation would be stronger and would be occurring at lower speeds in novice drivers. Effects were expected in position accuracy and cone hits as well as poor stabilization at the end of the maneuver in Lane 3. Such a poor stabilization is associated with loss of control and road departure accidents, and can be remedied by ESC. Table 8: Overview of the Test Conditions Session Speed 1 70 2 80 3 85 4 90 5 95 6 100 7 105 8 70 Figure 10: ISO Double Lane Change Maneuver 32 Figure 11: Adapted Double Lane Change Maneuver (Part A) 4.2.2 EXPERIMENT INSTRUCTIONS, PART B: HIGH SPEED CORNERING Part B of the experiment consisted of two sessions. In both the sessions participants were required to drive on a flat road with large corners (oval track). The track consisted of two straight road segments of 75 meters each and two large corners of 75-meter radius (Figure 12) and had a lane width of 6 meters. Session 1 required them to drive on a normal asphalt surface (µ=1) whereas Session 2 took place on a low friction surface (µ=0.4). It was hypothesized that experienced drivers would be more capable to sense the traction limits, select a curve speed approximating the maximum attainable speed, and accurately control the vehicle lateral position. Furthermore it was hypothesized that experienced drivers would be more capable to detect and assess the friction change and to adapt speed and steering control accordingly. The speed control in this part of the experiment was manual and the participants were asked to drive at maximum speed possible without losing control. Participants were only told that the friction of the road surface was reduced, but no information was given about the percentage reduction in the friction. For each session participants were given 12 minutes allowing multiple laps. Between the 2 sessions there was a 2-minute break during which they completed the NASA TLX questionnaire for measuring workload. The participant remained seated in the simulator during the break. Figure 12: Oval Test Track (Part B) 33 4.2.3 PARTICIPANTS The participants were divided into two groups based on their age. Group one represents the novice drivers (all male), aged from 18-21 years (Mean = 20.05, SD = 0.85) and consisted of 19 participants. Group two represented experienced drivers (all male), aged 25-35 years (Mean = 28.17, SD = 2.48) and consisted of 18 participants. All the participants were recruited from the Delft University of Technology campus. Before starting the experiment, the participants were asked to fill out a questionnaire regarding their driving experience and driving knowledge based on questions from the Driver Behavior Questionnaire (Reason et al., 1990) and Self-Assessment Questionnaire (McKenna et al., 1990) (see appendix 2 for the complete intake questionnaire). All participants provided written informed consent and the research was approved by the Human Research Ethics Committee of the Delft University of Technology (see appendix 1). 4.2.4 DEPENDANT MEASURES Each speed condition in experiment A had 5 sessions per driver and the mean of the dependant measures of the 5 sessions was taken. No sessions were excluded from the analysis. For experiment B first different laps were selected based on the vehicle position. Data for the two mentioned curves was then separated for each lap. The entry and exit of the curve was based on vehicle X-Y position (see Figure 13). Figure 13: X-Y Position of the selected curves for analysis The cases in which all the four wheels left the track were considered as “loss of control” or “road departure”. Laps with road departures were excluded from further analysis. For analysis the steering data is filtered using a low pass Butterworth filter (2nd order, 3Hz). All the other data was filtered using a Butterworth 2nd order 10Hz low pass filter. The performance of the drivers was analyzed using different metrics discussed in Table 8 below. Similar to the previous experiment (Chapter 3), Wilcoxon-Mann-Whitney test was used to deal with the non-normal distributions. 34 Table 9: Dependant measures used for analysis METRIC DESCRIPTION UNITS Root Mean Square Deviation Of The Lateral Acceleration (Part A) First the mean lateral acceleration is calculated for the maneuver. The start and stop point of the maneuver is predefined based on the vehicle position (X-Direction). Thus for all the drivers the maneuver length is same based on the distance travelled. RMSD from the mean of the lateral acceleration is then calculated through the 5 sessions of every speed. First the mean path taken by the driver while completing the maneuver is calculated at every speed. The start and stop point of the maneuver is predefined as explained above. Root mean square deviation from the mean path is then calculated for the 5 sessions of every speed. Root mean square deviation from the mid position of the lanes (defined by the cones) calculated for the 5 sessions of every speed. Vehicle position data and the cone position data is used to calculate the average number of cones hit by each driver and every speed Three main steering inputs are given by the driver to complete the double lane change maneuver. These three maximum angles are thus recorded (See Figure 14). Vehicle X-position when the three maximum angles mentioned above are reached is calculated. Part A: Average steering rate used to achieve the 3 maximum steering angles (as mentioned above) is calculated. Part B: Average steering wheel rate from the initial steering input (Steering angle>3) to the curve exit point Part A: Average rate of change of the steering wheel acceleration (jerk) used to achieve the 3 maximum steering angles (as mentioned above) is calculated. Part B: Average rate of change of the steering wheel acceleration (jerk) from the initial steering input (Steering angle>3) to the curve exit point Number or times the steering wheel input direction is changed with steering angle change < 3 degrees m/s2 Root Mean Square Deviation from the Mean Path (Part A) Root Mean Square Deviation from the Mid-Path (Part A) Average Number Of Cones Hit (Part A) Maximum Steer Angle (Part A) Position Of Steering Input (Part A) Average Steer Rate (Part A and Part B) Average Steer Jerk (Part A and Part B) Steering Reversal Rate (Part B) Mean Lateral Acceleration (Part B) Mean distance from inside of the curve (Part B) Deviation from inside of the curve (Part B) Deviation from the mean path (Part B) m m deg m deg/s deg/s3 reversal/ sec Mean of the lateral acceleration from the entry to the exit g point of the curve Mean distance of the vehicle from the inside of the curve from m the entry to the exit point of the curve Deviation of the vehicle position from the inside of the lane m from the entry to the exit point of the curve. A low value for this parameter means that the driver tries to maintain a constant distance from the inside of the curve. Deviation from the mean path chosen by the drivers in various m laps. 35 4.2.5 QUESTIONNAIRE Before the simulator experiment the participants were asked to fill in a questionnaire to judge their driving experience and driving knowledge. A part of the questionnaire required the participants to rate their ability to perform certain driving tasks out of a score of 10. They were also asked what they thought would be the rating of an average driver. The scores for the self-assessment and the average driver were subtracted and the overall sum for the entire 9-driving tasks was calculated (Table 10). A negative overall score meant that the driver assessed himself as better than an average driver, a score of 0 means that the driver put himself equal to an average driver and a positive score meant that the driver assesses himself as inferior to an average driver. All the experienced drivers rated themselves above average except one who rated himself equal to an average driver. 7 novice drivers rated themselves as above average, 1 driver rated himself equal to an average driver, and the rest put themselves below average. The participants were also asked vehicle dynamics related questions to judge their knowledge. The details of the questions can be seen in Appendix 6. DRIVING TASKS OVERALL SELF ASSESSMENT FOR THE 9- Table 10: Self Assessment score Mean (SD) Experienced -10 -1 -1 -15 -18 -13 -2 -3 -1 -8 -17 -14 -8 0 -8 -8 -5 -2 -7.11 (6.39) Novices 1 -4 20 -4 -16 -4 29 0 -1 8 3 6 -3 6 2 12 6 -4 1 3.05 (9.79) 36 4.3A RESULTS PART A: DOUBLE LANE CHANGE All the participants managed to perform the lane change at all speeds (from 70 km/h to 105 km/h). Increase in speed resulted in increase in the number of cones hit however the data was used for analysis as mentioned above irrespective of the number of cones hit. As explained in Table 9 above, the driver provides three main steering inputs while completing the double lane change maneuver as shown in Figure 14 below. Maneuver A-C: Steering input to the left to go from Lane 1 to Lane 2 Maneuver C-D: Steering input to the right to straighten the vehicle and to go from Lane 2 to Lane 3 Maneuver D-F: Steering input to the left for straightening the vehicle in Lane 3 Figure 14: Double lane change steering trace The maximum steering angle metrics were determined at points B, D and E. 4.3A.1 PATH FOLLOWED In the first step we analyze the path driven by the two groups of drivers. Figure 15 below shows the paths at increasing speeds from 70 to 105 km/h. As can be seen the experienced group shows better positioning of the vehicle in the middle of the lanes compared to the novices. Novices tend to be offset from the center position while entering the first lane. 37 38 Figure 15: Path followed during the Double Lane Change test (Black dots represent the cone position) This shows that the novice drivers are unable to judge the correct timing of the inputs and hence provide late inputs, which results in poor vehicle positioning as compared to the experienced drivers. This can be because the experienced drivers have a better and well-developed internal vehicle model and better perception skills, which enables them to judge the correct inputs and the timing. Moreover they are also better at anticipating the changes required in the control action on increasing the vehicle speed. In Figure 16 below the steering wheel angle as a function the vehicle position is shown. As can be seen the input of the novices always lags behind the input of the experienced drivers. Figure 16 also shows that novices try to correct the inputs by either giving higher steering angle or faster steering inputs. 39 Figure 16: Mean Steering Input for Experienced and Novice drivers 40 This can also be seen in the data in Table 11 below which shows the position at which maximum steering angles were achieved during the maneuver. The data shows (as also seen from Figure 16) that experienced drivers achieve the maximum steering angle earlier than novices. Table 11 shows that novices reach the maximum steering angle for the first input after Lane 1 exit whereas experienced drivers achieve the maximum angle before or at the exit of Lane 1. Table 11: Vehicle X-position when maximum steering input is given SPEED (km/h) LANE 1 70 80 85 90 95 100 105 EXPERIENCED 15.27 (2.70) 15.40 (3.08) 14.4 (3.52) 14.16 (3.84) 14.27 (3.32) 13.45 (3.44) 11.25 (4.68) NOVICES 20.30 (3.22) 19.82 (2.93) 19.14 (3.40) 19.10 (2.88) 17.55 (3.54) 17.83 (3.39) 17.85 (2.72) < 0.001 < 0.001 0.001 <0.001 0.008 <0.001 <0.001 66.45 (5.44) 63.23 (3.94) 62.55 (3.18) 61.60 (4.60) 59.20 (6.13) 62.04 (6.94) 58.03 (7.09) 65.38 (6.88) 0.573 66.84 (7.34) 0.079 66.92 (5.11) 0.007 67.01 (5.40) 0.004 64.81 (7.30) 0.043 69.87 (4.68) 0.001 69.11 (5.89) <0.001 94.69 (4.11) 97.38 (3.51) 95.46 (3.37) 97.09 (2.79) 96.40 (3.19) 96.56 (2.80) 98.57 (4.35) 95.67 (3.40) 0.455 97.42 (4.16) 0.820 97.95 (3.86) 0.070 96.88 (3.78) 0.337 98.33 (3.95) 0.110 99.87 (2.90) 0.002 101.67 (4.50) 0.043 p-value LANE 2 EXPERIENCED NOVICES p-value LANE 3 EXPERIENCED NOVICES p-value 4.3A.2 STRATEGY AND PERFORMANCE Figure 17 below shows the comparison between the experienced and the novice drivers based on the deviation of the root mean square acceleration achieved by the drivers while completing the task. As can be seen the experienced group shows lower deviation at all speeds compared to the novices. This indicates that the experienced drivers have a particular strategy, which they follow repeatedly, thereby showing less deviation. The repeatability in behavior can also be seen in terms of the root mean square deviation from the mean path followed by the drivers (Figure 17). The experienced drivers show much less deviation from the mean path compared to the novices. This can be because of better vehicle control skills, which enables them to execute their strategy repeatedly. There is an increase in the deviation with increase in speed for both the groups. (No clear effect of speed was seen on performance) 41 Figure 17: Repeatability in performance with respect to root mean square deviation in lateral acceleration and mean path (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) Deviation from the mid-position of the lane reveals the performance of the two groups. Figure 18 shows the experienced group has much smaller deviations from the mid path, which directly relates to better task performance. The same can also be seen in terms of the number of cones hit by the drivers (Figure 18). Figure 18: Performance in terms of root mean square deviation from the mid path and the number of cones hit (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) 42 Figure 19 shows the relation between the overall assessment score from Table 10 and the number of cones hit per session at different speed. 43 Figure 19: Self-Assessment versus actual performance (number of cones hit) Table 12 shows the Pearson correlation coefficients for the two groups at the different speed conditions. It can be seen that the assessment scores have a good correlation with performance for the experienced group. Novices on the other hand show very poor correlation between the two parameters. Table 12 clearly shows that the novices are poor at assessing themselves and often tend to overestimate their abilities. Experienced drivers on the other hand have a good self-assessment of their own abilities. Table 12: Correlation coefficient of Self-Assessment versus number of cones hit Experienced Novices Speed (km/h) 70 80 85 90 95 100 105 Correlation Coefficient 0.65 0.63 0.64 0.76 0.52 0.67 0.68 p-value Correlation Coefficient p-value 0.004 0.005 0.004 <0.001 0.028 0.002 0.002 0.22 0.05 0.17 0.09 0.09 0.03 0.11 0.357 0.830 0.486 0.705 0.718 0.914 0.647 44 4.3A.3 CONTROL STRATEGY As can be seen from Figure 20 below, the novices maintain higher steering rate at all speed conditions compared to the experienced group. The experienced group shows continuous increase in the rate with increase in speed. The novice drivers show no clear trend. The value of average steering jerk also shows the same trend with novices having higher values as compared to the experienced group (Figure 20). Figure 20: Comparison of Average steering rate and Average Steering Jerk between Novices and Experienced drivers (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) The maximum steering angles at three positions achieved in the maneuvers at different speed conditions are shown in Figure 21 below. It can be seen that the novices have lower steering angle for the first maneuver i.e. going from Lane 1 to Lane 2 at higher speeds (speed >85), but for the other two maneuvers the novices maintain a higher steering angle. As shown in Table 13, the input given by the novices is later compared to the experienced drivers. The combined effect of the lower and late inputs results in novices lagging behind in terms of input versus vehicle positioning. Thus in the second and third maneuver they try to compensate for this lag and hence have higher steering activity, which is shown in Figure 20 &21. 45 Figure 21: Maximum Steering Wheel Angle for the three steering inputs (See Figure 14). Top left represents the mean of the maximum steering angle in the first maneuver, top right represents the mean in the second maneuver and bottom represents the mean in the third maneuver (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) In the final steering maneuver it can be seen that novices try to compensate for the above-mentioned lag, showing higher steering activity than the experienced drivers. Figure 22 represents the steering maneuver for two representative drivers at 85km/h. As can be seen, steering input of the novice driver is lagging behind the experienced driver. During the final steering maneuver between positions C-E the novice driver tries to compensate for this lag and thus shows higher steering activity in terms of steering angle (position C and E), steering rate and steering jerk. 46 Figure 22: Steering input comparison between Experienced and Novice driver The results of the steering metrics (rate and jerk) are shown in Table 13. It can be seen that novices have higher steering activity compared to the experienced drivers. There is significant difference between the two groups as can be seen from the p-values. Table 13: Steering Rate and Steering jerk (Steering Input C-E) Speed Steering 80 85 90 95 100 105 46.38 55.98 60.23 65.73 66.99 67.56 (12.01) (11.93) (18.91) (17.46) (22.03) (13.70) (13.39) 60.78 73.04 69.51 76.01 73.85 86.39 84.03 (23.29) (33.52) (20.96) (27.15) (23.50) (29.87) (24.88) 0.035 0.003 0.040 0.062 0.218 0.032 0.008 Experienced 2498 2547 2440 2547 2740 2591 2809 (718) (761) (820) (865) (1029) (858) (928) 3278 3382 3124 3337 3198 3364 3776 (1435) (1351) (700) (900) (871) (810) (1658) 0.032 0.022 0.004 0.004 0.050 0.003 0.020 Experienced 46.28 Rate (rad/s) Novices p-value Steering Jerk (rad/s3) 70 Novices p-value Figure 23 shows the mean lateral acceleration for experienced and novice drivers. As can be seen, novices maintain a higher mean lateral acceleration as compared to the experienced drivers at all speeds, which could 47 possibly be attributed to higher steering activity shown by the novices especially in the later stages of the maneuver. Figure 23: Mean lateral Acceleration at all speeds (Bold dotted lines represent the mean whereas the upper and lower boundary represents the 25th and 75th percentile) 4.3A.4 TEST RE-TEST REPEATABILITY After finishing all the speed selections (70-105 km/h), the participants performed the double lane change test again at 70 km/h. This was done to study the learning among the drivers. It can be seen from Table 14 that the experienced and novices show no significant improvement in terms of deviation from the mean and mid path. Thus there is no significant improvement in task performance. Experienced drivers show a significant reduction in the values of the steering metrics of rate and jerk (p-value < 0.001). This indicates that the whole group shows the same trend. Similarly novices also show a significant reduction in the steering rate values (p-value = 0.003). Both the groups also advance their steering inputs significantly as can be seen from the lower value of steering position, which indicates the X-position at which the first steering input was given. Table 14: Change in performance for 70 km/h double lane change RMS_MEAN RMS_MID Steering rate Steering jerk Steer Position EXPERIENCED 1st Trial 2nd Trial p-value 0.16 (0.11) 0.10 (0.03) 0.059 0.28 (0.10) 0.26(0.06) 0.325 37.24 (9.88) 29.86 (13.78) <0.001 2289 (596) 1838 (518) <0.001 165.42 (2.70) 163.03 (3.87) 0.008 2782 (854) 2414 (932) 0.173 170.42 (3.26) 167.60 (3.11) <0.001 NOVICES st 1 Trial 2nd Trial p-value 0.24 (0.09) 0.22 (0.09) 0.382 0.59 (0.14) 0.54 (0.14) 0.089 48.27 (14.44) 36.25 (12.10) 0.003 48 4.3A.5 DISCUSSION As can be seen from the results presented in above sections, the experienced group shows better performance in the double lane change maneuver in terms of lower root mean square deviation from the mid path and lesser average number of cones hit. One of the reasons for the poor performance shown by the novices was the improper positioning of the vehicle while entering lane 1. The novices were always offset (right) of the mid-path, which could be evidence of poor perception of the vehicle width and lane width. It was hypothesized that task performance would degrade with increase in speed and that this degradation would be stronger and would occur at lower speeds in novice drivers. The experiment was unable to prove this hypothesis, as there was no clear degradation in performance of the novice or experienced drivers in terms of number of cones hit. Novices show poor repeatability in following their own strategy in terms of higher root mean square deviation from the mean path and mean lateral acceleration. Poor repeatability can be evidence that either the novices didn’t have any particular strategy and were using hit and trial method to get the best performance or they had a particular strategy but were not able to follow it due to poor vehicle control skills. As the speed of the maneuver was constant during each trial (controlled automatically) and the lane boundaries were defined, the double lane change test was effectively an open loop test. But the drivers showed some closed loop behavior towards the end of the maneuver to correct for the deviation in vehicle positioning. Unlike the results from the experiment explained in Chapter 3, the DLC test results show that the novice drivers have higher steering activity compared to the experienced drivers. As described in section 4.3A the DLC maneuver has three main steering inputs. It is hypothesized that the first and second steering inputs are more open loop whereas the third input is a more closed loop compensatory action. It is seen that the novices have a lower maximum value of first steering input at higher speeds (speed>85) as compared to experienced drivers. This can be indicative of poor perception on the width of the road and the distance between the lanes. It was also seen that novices tend to give late steering input in the initial stages of the maneuver as compared to the experienced drivers. Novices show inaccuracy in judging the correct timing of the control action and hence provide late initial steering input (similar to the experiment one result of experts having a more precise and accurate braking point). The novices seem to try to compensate for the delay in steering input in the later stages, thereby showing higher steering activity in terms of steering wheel angle, average steering rate and average steering jerk. Results also show that the novices maintained a higher mean lateral acceleration at all speeds as compared to experienced drivers. This can correlated to novices showing higher steering activity especially in the later stages of the maneuver. Data revealed that the novices show a decrease in the value of maximum steering angle for the first steering input with increase in session speed. Increase in speed (and hence required lateral acceleration) essentially requires higher slip angles and a larger steering wheel angle. The reason for this behavior might 49 be evidence that the novices have a poor internal vehicle model and hence are unable to update their control actions in changing driving scenarios. The increase in speed increases the speed of visual flow and hence novices might find in difficult to perceive and judge the required control inputs. Moreover maybe the novices are not comfortable in providing high steering inputs as they have rarely been in such situations and are unaware of how the vehicle will react to the control inputs. Experienced drivers on the other hand realize that the change in condition (here increasing speed) requires change in control actions and hence adapt their inputs accordingly. The results also show that the novices are poor at self-assessment compared to the experienced drivers. Performance versus self-assessment showed strong correlation for the experienced group. Comparing the first and last session at 70 km/h no significant learning was seen in any of the groups in terms of improvement in performance although significant reduction in steering metric values was shown within the groups. It can be concluded from the results that steering metrics of timing/position of steering input, average steering rate and average steering jerk can be used to differentiate novice and experienced group of drivers, and are apparently more discriminative than performance based metrics. The significant effects of learning indicate that effects of increasing speed might be confounded by effects of learning, but as stated above limited effects of speed were observed. Thus the study shows correlation between experience and performance with experienced drivers performing better than the novices. Late and insufficient steering inputs and higher steering activity separated the novice drivers from the experienced group. The data also shows that steering rate, steering jerk and position of steering input metrics can be used to differentiate novice and experienced drivers. 4.3B RESULTS PART B: CORNERING During Session 1 (normal friction road surface), experienced drivers had 104/400 (26%) road departures whereas novices had 131/364 (36%) road departures. The reason behind the higher percentage among novices could be that they take more time to adapt to the simulator environment as compared to experienced drivers. During Session 2 (low friction surface) the amount of road departures was 89/281 (31%) for the experienced drivers and 98/279 (35%) for novices. 4.3B.1 CURVE-TIMES AND LATERAL ACCELERATION Experienced drivers have lower curve times and maintain a higher lateral acceleration while going around the curves compared to novices (Table 15 & 16). Drivers were closer to the traction limit in Run 2 (0.2g0.3g) than in Run 1 (0.4g-0.5g). Experts were faster than novices by a margin of 1.7 seconds in Session 1 and by 1-1.5 seconds in Session 2. The two groups perform significantly different as can be seen from the pvalues indicated in Table 15 & 16 below. 50 Table 15: Curve-Times (sec) Curve 1 Curve 2 Session1 (Normal Friction) Experienced Novices p-value 13.26(1.02) 14.94 (1.32) <0.001 13.14 (1.04) 14.89 (1.26) <0.001 Session2 (Low Friction) Experienced Novices p-value 18.95 (1.07) 20.45 (1.31) <0.001 19.10 (1.31) 20.21 (1.06) 0.011 Table 16: Mean Lateral Acceleration (g) Curve 1 Curve 2 Session1 (Normal Friction) Experienced Novices p-value 0.49 (0.13) 0.42 (0.07) 0.022 0.47 (0.11) 0.42 (0.07) 0.043 Session2 (Low Friction) Experienced Novices p-value 0.29 (0.06) 0.25(0.05) 0.037 0.29 (0.05 0.25 (0.04) 0.004 4.3B.2 STEERING PERFORMANCE Table 17 shows that experienced drivers have higher steering activity in terms of steering jerk and steering reversal rate. The two groups show significant difference in steering jerk and steering reversal rate for curve 1 and curve 2 in both the sessions. Steering rate showed no significant difference between the groups for Session1. In Session 2 experienced drivers showed significantly higher steering rate values as compared to the novices. Results are similar to what was found in the first experiment (Chapter 3). In Session 2 there is a reduction in steering metric values, which is because of low coefficient of friction of the road. Thus the two groups adapted to the situation by reducing the speed of the vehicle and reducing the magnitude of control inputs. The percentage road departures are approximately same for the two groups as discussed earlier. 51 Table 17: Steering Metrics Experienced Novices Steer CURVE 1 Steer Rate Jerk 28.71 jerk Session1 (Normal Friction) 2593 1.20 28.40 2570 1.19 SRR (348) (0.16) (3.91) (290) (0.14) 32.56 2320 0.98 32.71 2357 1.01 (10.88) (400) (0.18) (10.06) (442) (0.19) 0.010 0.001 0.176 Session2 (Low Friction) 2490 0.88 25.46 0.032 0.003 2596 0.88 (5.97) (376) (0.20) (8.52) (575) (0.21) 18.39 2121 0.78 20.28 2139 0.81 (5.00) (209) (0.18) (4.72) (152) (0.15) 0.037 0.002 0.111 0.043 <0.001 0.308 0.294 Experienced 22.7 p-value Rate SRR (5.65) p-value Novices Steer CURVE 2 Steer As seen in Table 17, both group of drivers show differences in steering behavior. To understand the reason behind the difference in steering strategy we analyze the under-steer/over-steer behavior of the vehicle. Under-steer/over-steer coefficient is calculated using the front and rear slip angles values. Table 18 shows that for both the curves in session 1, half the drivers from the experienced group drive the vehicle in an oversteer state and half in under-steer state. The novices on the other hand show more over-steer behavior as compared to the experienced group (Table 18). The percentage of under-steer driving shows significant difference between the two groups (p<0.05). In session 2, the experienced drivers adopt a more over-steer strategy as compared to the novices. The difference is significant in curve 2 of session 2 (p<0.05). Table 18: Percentage Under-steer Session1 (Normal Friction) Experienced Novices p-value Curve 1 % Under-steer 50.99 (10.05) 37.93 (10.59) 0.001 Curve 2 % Under-steer 52.35 (13.59) 40.12 (15.64) 0.017 Session2 (Low Friction) Experienced Novices p-value 38.04 (11.46) 42.01 (12.80) 0.616 34.34 (11.97) 43.60 (10.71) 0.018 52 4.3B.3 PATH STRATEGY Table 19 below describes the path strategy followed by experienced and novice group of drivers while taking the corner. The first column of Table 19 indicates the mean distance of the vehicle from the inside of the curve (D1). Column 2 represents the deviation of the vehicle position from the inside of the lane (D2). A low value for this parameter means that the driver tries to maintain a constant distance from the inside of the curve. Column 3 represents the deviation from the mean path (D3) chosen by the drivers in various laps. This parameter represents the repeatability of drivers in following their own strategy. As can be seen experienced and novices follow the same strategy and have a mean distance of 1.92 and 1.95 meters during Session 1 and 1.85 and 2.0 meters during Session 2. The deviation from the inside of the curve is also same for both groups of drivers in the range of 0.51 and 0.59 meters for Session 1 and 0.71 and 0.67 meters for Session 2. Table 19: Path Strategy (D1: Mean distance from inside of the curve; D2: Deviation from inside of the curve; D3: Deviation from the mean path Session1 D1 CURVE 1 D2 D3 D1 CURVE 2 D2 D3 Experienced 1.94 (0.39) 1.93 (0.42) 0.704 0.50 (0.14) 0.58 (0.16) 0.207 0.52 (0.18) 0.70 (0.18) 0.005 2.07 (0.38) 1.89 (0.43) 0.150 0.50 (0.07) 0.57 (0.16) 0.207 0.53 (0.15) 0.67 (0.17) 0.007 2.04 (0.40) 1.95 (0.62) 0.595 0.80 (0.42) 0.63 (0.31) 0.149 1.25 (0.47) 0.88 (0.56) 0.019 1.85 (0.45) 2.12 (0.58) 0.066 0.64 (0.29) 0.60 (0.22) 0.796 0.93 (0.41) 1.06 (0.67) 0.750 Novices p-value Session2 Experienced Novices p-value Table 19 shows that the two groups maintain approximately the same mean distance from the inside of the curve. To see if there is any difference in the path chosen between the two groups we analyze the vehicle positioning by the drivers while taking the corner. For this the driver with the mean performance and the best performance based on the lap-times were selected. Figure 24 below shows the path chosen by experienced drivers. As can be seen there is no particular strategy that can be related to the particular group. Two types of path strategies were found in both groups. The first strategy relates to the driver entering the curve a constant distance from the inside of the curve and then moving towards the outside of the curve while cornering and exiting. This can be seen in the mean performance of experienced group in Figure 24 (bottom left). Most of the novice drivers tried to follow this strategy. In the second strategy the driver enters from the outside of the 53 curve, while cornering goes close to the inside of the curve and then moves towards the outside of the curve while exiting, thus following a racing line. This can be seen in the best performance of experienced and novice group in Figure 24 (middle left and right). Most of the experienced drivers tried to follow this strategy. In both the groups, best performance in terms of lap-time was achieved follow the second strategy. Best performance shown by experienced driver in Figure 24 had a mean curve-time of 12.19 seconds which is 1.07 seconds faster than the mean of the group. Best performance shown by novice driver in Figure 24 had a mean curve-time of 13.51 seconds which is 1.43 seconds faster than the mean of the group. Mean performance for an experienced driver shows both the abovementioned strategies, whereas for the novices no particular repeatable strategy can be seen in Figure 24 (bottom right). Figure 24: Deviation from inside of the curve: Top (left for novice and right for expert) is the path followed; Middle left is the best performance by experienced driver; Middle right is the best performance by novice driver; Bottom left is the mean performance by experienced driver; Middle right is the mean performance by novice driver 54 Moreover, novice drivers were poor at following any particular strategy, as can be seen from Table 19 above. Parameter D3 that gives the deviation from the mean path shows significant difference between the two groups with experienced drives showing lower deviation compared to the novices in Session 1. The deviation is higher for the experienced group on the low friction surface in Session 2, maybe because experienced drivers tried to achieve the better lap-times and hence drove the vehicle at higher lateral acceleration, whereas the novices’ main aim was to keep the vehicle on track. 4.3B.4 DISCUSSION A high speed cornering experiment was performed to quantify the differences in performance and strategy between novice and experienced drivers. Results showed better performance for experienced drivers in terms of lower lap-times and higher average lateral acceleration as compared to the novices. The first experiment session (normal friction road surface) showed more road departures for novices as compared to the experienced drivers. This could possibly be because novices take time to adapt to driving in a simulator environment and understanding test procedure. It can also be possible that the novices took a longer time to judge the correct maximum speed and appropriate control inputs in order to go around the track without road departures, which can be evidence of lower vehicle dynamics and response knowledge. The results also show that there is a difference in steering behavior between the two groups. Expert drivers show higher steering activity in terms of steering rate, steering jerk and steering reversal rate. These results are consistent with the racing experiment discussed in chapter 3 and the literature study discussed in chapter 2, which showed higher steering activity among expert drivers in terms of steering wheel angle, average steering jerk and frequency of steering inputs. As discussed in chapter 3, the driving task is a combination of feed-forward and feedback control. High steering activity shown by the experienced drivers could possibly be evidence of higher feed-forward and feedback gain as compared to the novices. Possibly the lower steering activity shown by the novices can be attributed to inaccurate perception of the road curvature, hence imprecise judgment of the required control inputs. Experts might be better at perceiving the information from the surroundings and might have a better vehicle dynamics knowledge, allowing them to better judge the correct control inputs. No single path strategy was found to define the group behavior for the two groups. The path strategy of entering from the outside of the curve, going close to the inside of the curve while cornering and then moving towards the outside of the curve while exiting showed the best performance for both the groups in terms of curve-times. This strategy is similar to the one shown by the experts in chapter 3 and by Treffner (2002). Experienced drivers showed better repeatability in following their strategy in terms of deviation from their mean path as compared to the novices. Poor repeatability can be evidence that either the novices didn’t have any particular path strategy and were using hit and trial method to get the best performance or they had 55 a particular strategy but were not able to follow it due to poor vehicle control skills. The data also revealed that the novices drive in an over-steer condition in both the sessions whereas the experienced drivers drove in under-steer state during session 1 but shifted to a more over-steer strategy in session 2. This could possibly be attributed to novices braking while cornering hence inducing over-steer in the vehicle during session 1. In session 2 the experienced drivers were close to the traction limit and hence probably applied an over-steer strategy to control the vehicle. Summarizing, it can be concluded that the experienced drivers performed better than the novices in the highspeed cornering task. Higher steering activity shown by the experienced drivers was possibly the reason better performance. Differences in driving strategy and repeatability in following the path strategy most clearly separated the experienced from the novice drivers. Metrics like steering rate, steering jerk and steering reversal rate used to differentiate the two groups of drivers showed major and significant differences. 56 CHAPTER 5: CONCLUSION AND FUTURE RESEARCH 5.1 CONCLUSION The present study was aimed at finding differences in driver behavior and performance between expert, normal (experienced) and novice drivers in different driving situations. Three experiments were performed to analyze these differences, one in a racing environment and two in a standardized driving environment. Performance of expert and normal drivers was analyzed in a high-speed driving task using a dataset from a previously conducted racing experiment. The second experiment (referred as Part A in Chapter 4) compared experienced versus novice drivers in a double lane change (DLC) maneuver. The third experiment (referred as Part B in Chapter 4) compared experienced versus novice drivers in a high-speed cornering task on an oval track on two road friction conditions. The three experiments showed strong and significant differences between the different groups of drivers in terms of performance, steering behavior, path strategy and repeatability. In the racing experiment experts showed better performance in terms of lower lap-times and higher average lateral acceleration as compared to the normal drivers. Similarly in the high-speed cornering task experienced drivers were faster and maintained a higher lateral acceleration as compared to the novices. In the DLC test experienced driver showed better performance in terms of lesser deviation from the mid-path and lower average number of cones hit as compared to the novices. Thus the three experiments revealed clear correlation between driver experience and performance. The experiments also showed clear differences between the three groups in terms of steering behavior. The racing experiment and high-speed cornering experiment showed that better performance was achieved using higher steering activity. In the racing experiment expert drivers showed significantly higher steering activity in terms of steering rate and steering jerk whereas in the high speed cornering experiment it was seen that experienced drivers showed significantly higher steering activity in terms of steering jerk and steering reversal rate which resulted in better performance by the two groups in the individual experiments. The DLC test results showed that the novice drivers have higher steering activity and lateral acceleration compared to the experienced drivers. This is opposite to what was found in the two other experiments. On detailed analysis it was seen that novices tend to give late steering input in the initial stages of the maneuver. Moreover the magnitude of the maximum steering angle of the first steering input was lower for the novices as compared to the experienced drivers and there was a decrease in the magnitude with increase in speed. Increase in speed (hence lateral acceleration) essentially requires higher slip angle and hence initial higher steering angle, which was not generated by the novices. 57 The experiments also revealed differences in the adopted strategy and repeatability in following this strategy between the different groups of drivers. In the racing experiment, while cornering, normal drivers maintained a constant distance from the inside of the corner while the experts positioned their vehicle towards the outside of the corner while entering and exiting and went close to the inside of the curve in the mid-section (trying to follow the race line). In the high-speed cornering experiment no particular strategy was indicative of group behavior but it was found that experienced and novice drivers achieved best performance while following a similar strategy as the experts in the racing experiment. At the end of the DLC test, participants were asked to perform a re-test at 70km/h to analyze the learning among the drivers in the experiment. No significant improvement was shown in performance by any of the groups in the re-test although significant reduction was seen in the steering metric values for both the groups. In the racing experiment experts show precise timing and repeatability in control action in terms of the braking point. Experts brake at the same point across all the sessions as compared to normal drivers who show more deviation in their braking point. In the high speed cornering experiment, experienced drivers showed significantly less deviation from their chosen path as compared to the novices. Similarly in the DLC test experienced drivers showed consistency in following their strategy as compared to the novices in terms of lesser root mean square deviation from their chosen mean path and lower root mean square deviation in lateral acceleration. The results discussed above are consistent with previous research as discussed in chapter 2 which showed higher steering activity among expert drivers in terms of steering wheel angle, average steering jerk and frequency of steering inputs (Hollopeter, 2011; Zhang et al., 2009). Better repeatability and path strategy adopted by the experts and experienced drivers are similar to the results shown by Treffner et al. (2002). The driving task is a combination of open loop (feed-forward) and closed loop (feedback) control. In the racing experiment expert drivers and in the high speed cornering experiment the experienced drivers showed significantly higher steering activity in terms of steering rate, steering jerk and steering reversal rate. This can be attributed to higher feed-forward and feedback gain for the experts compared to the normal drivers and experienced drivers compared to the novices in the individual experiments. Thus it is possible that the experts use higher steering activity as corrective measure to achieve the optimal path. In the racing experiment it is possible that the higher steering activity shown by the experts is to stabilize the vehicle and keep it at the traction limits. Overall higher steering activity could be evidence of a better internal vehicle model, which means that the driver has a good knowledge of the vehicle dynamic response and control gained through experience. Compared to the normal drivers and novices, the expert and experienced drivers have a better understanding of what the current situation demands, which control inputs are required, and also how the vehicle will respond to the given control inputs. The promptness in steering action and the 58 ability to provide faster inputs could be the result of experience and practice. The DLC test results show that the novice drivers have higher steering activity compared to the experienced drivers. On detailed analysis it was seen that novices had a lower maximum value of first steering input at higher speeds (speed>85) as compared to experienced drivers, which is indicating poor anticipation of the required steering input. It was also seen that novices tend to give late steering inputs in the initial stages of the maneuver as compared to the experienced drivers. Novices show inaccuracy in judging the correct timing of the control action and hence providing late initial steering input possibly because of poor perception of the vehicle position and distance between the two lanes. The novices try to compensate for the lag in steering input and in the later stages of the DLC, showing higher steering activity in terms of steering wheel angle, average steering rate and average steering jerk. There was also a decrease in the value of maximum steering angle for the first steering maneuver with increase in speed for the novices, which can be evidence that novices have a poor internal vehicle model and hence are unable to update their control actions in changing driving scenarios. Maybe the novices were not comfortable in providing high steering inputs as they have rarely been in such situations and are unaware of how the vehicle will react to the high steering inputs. Experienced drivers on the other hand realize that the change in condition (increasing DLC speed) requires change in control actions and hence adapt their inputs accordingly. Repeatability in performance shown by the experts in the racing experiment and experienced drivers in the high speed cornering test can be attributed to better perception of the road curvature which results in a more accurate judgment of the speed at which to take the corner. A well-developed internal vehicle model enables them predicting the correct timing of braking to achieve the desired speed and superior vehicle control skills results in better repeatability in following the strategy. Poor repeatability in following the chosen strategy shown by the novices in the high speed cornering experiment as well as the DLC test can be evidence that either the novices didn’t have any particular strategy and were using a hit and trial method to get the best performance, which is indicative of poor vehicle dynamics and control knowledge or they had a particular strategy but were not able to follow it due to poor vehicle control skills. The DLC test results also showed that the novices were poor at self-assessment compared to the experienced drivers. Thus as novice group showed limited correlation, it is advisable to use measured metrics in order to classify drivers based on performance versus skill. Concluding, the experiments revealed that the three groups, experts, experienced and novices provide marked and strong differences in performance and strategy. There was a clear correlation between experience and performance with experts having better performance in the racing experiment as compared to the normal drivers and experienced drivers showing better performance in the high speed cornering and DLC test as compared to the novices. Performance metrics like lap-times and lateral acceleration showed 59 significant differences between groups in the racing experiment and the high speed cornering experiment whereas deviation from mid-path and average number of cones hit showed high significance in the DLC test. Clear and major differences were seen in the steering behavior between the groups. Higher steering activity was found to be the main reason for better performance in the racing experiment and the high speed cornering experiment. Appropriate control inputs (steering angle) and accurate timing of steering input resulted in better performance in the DLC test. Steering metrics like steering jerk, steering rate, steering reversal rate and timing (position) of steering input showed significant differences between the groups. Differences in path strategy and consistency in following the strategy also showed significant difference between the two groups. 5.2 FUTURE RESEARCH The experiments performed in this study quantified differences between experts, experienced and novice drivers with emphasis on steering behavior. The experiments performed showed higher steering activity resulting in better performance in cornering tasks and correct timing of control inputs resulted in better performance in the lane change task. Thus the results showed differences in performance based on skill and experience and that steering control metrics can be used to differentiate different drivers. Hence the optimal intervention of ADAS systems can be different for different drivers based on their skill level. The present study was focused on differentiating drivers in extreme steering conditions so that the results can be used in designing as driver adaptive ESC systems, which operate at vehicle limits. Although the results showed major and significant differences in quantifying the differences between the drivers but taking into account driver variability in ESC system design requires more research. Future research can focus on validating and analyzing the difference in optimal performance of a driver adaptable ESC system in extreme conditions i.e. for worst and best drivers. Further research is also required in finding ways to judge the driver performance in real time (i.e. while driving) in order to update the ESC system continuously for driver variability. Moreover the tests performed in this study were focused on classifying drivers based on steering behavior in controlled experiments. Research has to be done in order to correlate skill based on steering performance with skill based on real life driving performance. The results from the study can also be used to include driver variability into driver models for computer simulation. These models can be used for designing and testing driver adaptive ADAS systems Moreover the experiments were simulator-based and thus it would also be relevant to compare real vehicle testing results with the simulator results, as different perceptional cues (e.g. vestibular cues) were not present in the driving simulator. The present study was aimed at finding differences in driver behavior and performance comparing expert to normal (experienced) drivers and normal to novice drivers. Driving proficiency is expected to increase from novice, to normal to expert drivers, but this may well involve complex multidimensional learning processes. 60 Further effects on behavior and performance can be expected from gender, aging and physical and cognitive disorders. A further exploration of variation in steering behavior could apply the tests and metrics developed in this thesis to a much wider population. 61 REFERENCES 1. Brookhuis, K.A., Smiley, A. (1987). Alcohol, drugs and traffic safety. Road users and traffic safety, J.A. Rothengatter & R.A. de Bruin (Eds.), Assen: Van Gorcum, 83-105. 2. Cacciabue, Carlo P. & Carsten, Oliver. (2010). A simple model of driver behaviour to sustain design and safety assessment of automated systems in automotive environments. Applied ergonomics, 41. Retreived from http://www.biomedsearch.com/nih/simple-model-driver-behaviour-to/19450791.html 3. Erséus , A., Agebro, M. and Stensson Trigell A. (2007). Study of path tracking skill and strategy using a moving base simulator. FISITA TRANSACTIONS, paper F2006D075T. 4. Erséus , A., Stensson Trigell, A. and Drugge, L. (2008) A path tracking scenario without preview for analysis of driver characteristics. Proceedings of the 9th International Symposium on Advanced Vehicle Control (AVEC’08), Kobe, Japan. 5. Erséus A. (2010). PhD dissertation: Driver-Vehicle Interaction - Identification, Characterization and Modelling of Path Tracking Skill. Doctoral Thesis in Vehicle Dynamics, Royal Institute of Technology, TRITA-AVE 2010:29. Project cooperation with Saab Automobile and GM R&D. 6. Erséus, A. Stensson Trigell, A. and Drugge, L. (2010). Characteristics of path tracking skill on a curving road, submitted for publication. (Submitted for publication). Royal Institute of Technology, Stockholm, Sweden. 7. Erséus, A., Stensson Trigell, A. and Drugge, L. (2009). Methodology for finding parameters related to path tracking skill applied on a DLC-test in a moving base driving simulator. (Submitted for publication). Royal Institute of Technology, Stockholm, Sweden. 8. Erséus, A., Stensson Trigell, A. and Drugge, L. (2011). A path tracking driver model with representation of driving skill. International Journal Vehicle Systems Modeling and Testing, Vol. 6, No. 2, 2011. 9. Fildes B. N. and Lee S. J. (1993). The Speed Review: Road Environment, Behavior, Speed Limits, Enforcement and Crashes. Report No. CR 127, Federal Office of Road Safety, Canberra, Australia, September 1993. 10. Fuller R. (1984). A conceptualization of driving behavior as threat avoidance. Ergonomics, Volume 27(11), Nov 1984, 1139-1155. doi:10.1080/00140138408963596 11. Fuller R. (2000). The task capability interface model of the driving process. Journal Recherche Transports Securite (RTS) 66, 47-59. 12. Fuller R. (2005). Towards a general theory of driver behavior. Accident Analysis and Prevention, 37 461-472. 13. Gugerty, L. (1997, in press). Situation awareness in driving. In Handbook for Driving Simulation in Engineering, Medicine and Psychology. Eds.: J. Lee, M. Rizzo, D. Fisher & J. Caird. CRC Press. 62 14. Hollopeter, Joy N. (2011). Response of novice and experienced drivers to lateral control intervention to prevent lane departures. Master's thesis, University of Iowa. Retrieved from http://ir.uiowa.edu/etd/984. 15. Katzourakis, D., Velenis, E., Happee, R. (2011). Driver Control Actions in High-Speed Circular Driving. Proceedings of the Sixth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design. 16. Kiencke, U. and Nielsen, L. Automotive control systems: for engine, driveline, and vehicle, SpringerVerlag, Berlin, 2000. 17. Malik, Husnain (2011) An assessment system for evaluation of driving competencies. PhD thesis, Queensland University of Technology. Retrieved from http://eprints.qut.edu.au/view/person/Malik,.Husnain.html 18. Miller, B.W.; Chung Hee Hwang; Torkkola, K.; Massey, N. (2003). An Architecture for an intelligent driver assistance system. Intelligent Vehicles Symposium, 2003. Proceedings. IEEE, vol., no., pp. 639- 644, 9-11. doi: 10.1109/IVS.2003.1212987 19. Rizzi, M. (2005). Steering Behaviour of 44 Drivers in Lane Change Manoeuvres on a Slippery Surface. Master’s Thesis, Linköpings universitet: Institutionen för teknik och naturvetenskap. 20. Treffner, Paul & Barrett, Rod & Petersen, Andrew. (2002). Stability and skill in driving. Human movement science, 21. Retreived from http://www.biomedsearch.com/nih/Stability-skill-in- driving/12620719.html 21. Underwood, Csessiondall, & Chapman. (2003). Visual attention while driving: Sequences of eye fixations made by experienced and novice drivers. Ergonomics, 46, 629-646 22. Weiner E. L. & Curry R. E. (1980). Fight-deck automation: promises and problems. Ergonomics, 23(10), 995-1011. 23. Zhang, Y., Lin, W.C., Chin, Y.S. (2008). Driving Skill Characterization: A Feasibility Study. IEEE International Conference on Robotics and Automation, USA. 63 APPENDICES Appendix 1. Consent form Consent to Participate in a Driving Simulator Study Researchers: Naman Singh Negi (MSc student) P.M. van Leeuwen (PhD student; supervisor) Dr. Ir. R. Happee (supervisor) Email: [email protected] p.m.vanleeuwen @ tudelft.nl [email protected] Title of this study: Driver skill evaluation and classification in extreme steering conditions Location of the experiment: TU Delft. Faculty Mechanical, Maritime and Materials Engineering (3mE) Biomechanical Engineering Department (BMechE) X-Car Driving Simulator Mekelweg2, 2628CE, Delft Introduction: Before agreeing to participate in this study, it is important that the following explanation of the proposed procedures be read and understood. This document describes the purpose, procedures, benefits, risks, and possible discomforts of the study. It also describes the right to withdraw from the study at any time. Purpose of the study: The purpose of this driving-simulator study is to investigate driving behavior and skill level based on the driver control inputs, path strategy and adaptability to changing conditions. You will be 1 of approximately 20 participants taking part in this study. The results of this experiment will be statistically analyzed and published in a Master’s thesis and scientific publication. Duration: Your participation in this study will last approximately one hour. Procedures: Before driving in the simulator, you will be asked to fill out a questionnaire regarding your driving experience and driving knowledge. Next, you will seat in the driving simulator and you will be briefed in how to operate it. You can control the car using the steering wheel, the accelerator and brake pedal. The gearbox is in all cases set as automatic; you do not have to use the clutch and gearshift. The experiment consists of 2 parts. Between the two parts of the experiment, you will have a short break of 5 minutes in which you will be asked to answer a questionnaire outside the simulator about your physical and mental workload, discomfort in the previous session. In the first part of the experiment, you are supposed to make a double lane change maneuver. The schematic of 64 the maneuver is shown in the figure below. You are expected to keep the vehicle within the lane boundaries (depicted by the cones). Before starting the final trials a practice session of 5 minutes will be given to you to get accustomed to the simulator environment and the test conditions. During the final trials the speed control for this will be automatic and thus you can only control the car using the steering wheel. The experiment will start from a speed of 50km/h. You will be given 5 attempts at every speed selection. After every speed selection a 1-minute break will be given. 1 successful attempt out of the 5 is required to move to the next higher speed. Hitting of the cones, road departures and spin out (complete loss of control with the vehicle doing circles) will be considered as unsuccessful events. Speed increment will be in steps of 5km/h to 10km/h. FIGURE 1: DOUBLE LANE CHANGE MANEUVER Part 2 of the experiment consists of 2 sessions. In both the sessions you are required to drive on a flat road with large corners. The speed selection in this part of the experiment is manual. You have to drive around the road at maximum speed possible without losing control. Session 1 requires you to drive on a normal asphalt surface whereas in Session 2 you will be driving on a low friction surface. For each session you will be given 15 minutes. Between the 2 sessions there will be a 2 minute break. Risks, and discomforts: For some people, simulators and virtual environments may cause different type of sickness: visuomotor dysfunctions (such as eyestrain, blurred vision, difficulty in focusing), nausea, drowsiness, fatigue, or headache. These symptoms are similar to motion sickness. If you feel uncomfortable in any way, you are advised to stop the experiment. You can stop participating at any time without any negative consequences. If you do not feel well, then please take sufficient rest before leaving the laboratory. Confidentiality: All the data collected in this study will be kept confidential and will be used for research purposes only. Throughout the study you will be identified by a subject number only. Right to refuse or withdraw: Your participation is strictly voluntary and you may refuse to participate, or discontinue your participation at any time, without negative consequences. The investigator has the right to withdraw you from the study at any time for reasons related solely to you (for example, not following study-related directions from the investigator). Questions: If you have any questions concerning this study, you may contact Naman S. Negi ([email protected]), P.M. van Leeuwen (p.m.vanleeuwen @ tudelft.nl) or Dr. Ir. R. Happee ([email protected]) I have read and understood the information provided above. I give permission to process the data for the purposes described above. I voluntarily agree to participate in this study. Name: _____________________________________________ Signature of participant Date _________________ 65 Appendix 2: Questionnaire NAME AGE SEX 1. HOW MANY KILOMETERS DO YOU DRIVE EVERY YEAR A) LESS THAN 5000 B) 5000-10000 KM C) MORE THAN 10000 KM 2. ARE YOU A RACING ENTHUSIAST A) NO INTEREST B) SLIGHT INTEREST C) INTERESTED D) HARDCORE FAN 3. DO YOU PLAY CAR RELATED VIDEO GAMES A) NEVER B) SOMETIMES C) OFTEN D) ALMOST EVERYDAY 4. DO YOU HAVE KNOWLEDGE OF VEHICLE DYNAMICS A) NOT AT ALL B) BASIC KNOWLEDGE C) GOOD KNOWLEDGE D) EXPERT 5. HAVE YOU DRIVEN IN A SIMULATOR BEFORE A) NEVER B) FEW TIMES C) OFTEN D) REGULARLY 6. RATE YOURSELF AND AN AVERAGE DRIVER ON A SCALE OF 10 (0 (VERY POOR)- 10 (VERY GOOD)) IN THE FOLLOWING DRIVING SKILLS 1. 2. 3. 4. 5. 6. 7. 8. 9. OVERTAKING CHANGING LANES CORNERING JUDGING STOPPING DISTANCE JUDGING SPEED FOR BENDS/CORNERS JUDGING WIDTH OF VEHICLE ATTENTION TO OTHER VEHICLES PAYING ATTENTION TO ROAD SIGNS/PEDESTRIANS/CYCLISTS ETC CHANGING DRIVING TO SUIT WET/ICY/FOGGY CONDITIONS YOURSELF /10 /10 /10 /10 /10 /10 /10 AVERAGE DRIVER /10 /10 /10 /10 /10 /10 /10 /10 /10 /10 66 7. WHAT WILL HAPPEN IF THE DRIVER SUDDENLY BRAKES WHILE TAKING A CORNER? A) VEHICLE WILL SAFELY COME TO A STOP B) REAR WHEELS START TO SKID C) FRONT WHEELS START TO SKID 8. WHILE TAKING A RIGHT HAND CORNER THE DRIVER NOTICES THAT REAR OF THE VEHICLE IS SKIDDING OUT. WHICH OF THE FOLLOWING ACTIONS SHOULD HE TAKE? A) BRAKE B) INCREASE THROTTLE C) STEER INTO THE CORNER D) STEER OUT OF THE CORNER 9. THE DRIVER SUDDENLY BRAKES ON A STRAIGHT ROAD WITH LEFT SIDE OF THE CAR ON WATER. WHAT WILL HAPPEN? A) VEHICLE STOPS IN A STRAIGHT POSITION B) VEHICLE STEERS TO THE RIGHT C) VEHICLE STEERS TO THE LEFT. 10. DRAW THE PATH THAT YOU WOULD FOLLOW TO TAKE THE CORNER IN THE SHORTEST POSSIBLE TIME. 67 Appendix 3: Crashes Analysis We now analyze the laps in which the drivers crashed. The graph below shows that the crashes occur as the vehicle leave out of the track, not into the track. To investigate the causes we study the steering, brake and throttle data. We first compare the speeds of the drivers in crash and no-crash situation. It is possible that most of the crashes are because of over speeding. The table below shows the maximum speed of experts and novices while entering the corner. As can be seen from the data the expert drivers crash mainly because of over speeding as they try to push the car to the limits. On the other hand the non-experts have crashes at lower speeds also. The crashes must then be a result of improper steering, brake or throttle input or a combination of these inputs. SESSION 4 Expert NonExpert NO-CRASH Curve Curve 3 2 167.6 155.06 9 (4.9) (6.52) 169.61 151.2 153.24 5 2 (2.71) (8.32) (3.42) Curve 1 180.29 (1.52) Curve 1 180.8 1 (0.78) 168.1 8 (6.577 ) CRASH Curve 2 Curve 3 168.86 (7.76) 156.17 (4.27) 150.52 (9.8) 141.17 (17.28) In the figures below we compare the steeirng and braking inputs of the crash laps with the best lap (fastest) to understand the differences in the inputs and find the cause of loss of control. Dotted lines:Best lap Bold line: Crash laps Blue lines: Steering Angle Green Lines: Braking Red Lines: Path followed Black Lines: Speed of the vehcile Non-Expert versus the best lap 1) Late steering input given by novices. Expert gives the steering input much earlier and maintains a constant steering angle to negotiate the curve. (Figure 1) Figure 25: Late Steering Input 68 2) Higher and erratic braking shown by the novices. Not much use of the brakes by the expert once the constant steering angle is reached. (Figure 2) Figure 26: Erratic Braking 3) Over-steer controllability: The experts are able to control the over-steer generated as they go into the curve by giving a high steering wheel angle which makes the front tire slip more (thereby generating under-steer). The normal because of the as they go into the curve the steering and braking generates an over-steer in the vehicle, which they are unable to control. (Figure 3) Figure 27: Over-steer controllability 69 Expert versus the best lap 1) Late steering input given by the expert when they crash represented by the blues lines can be seen in Figure 4. Figure 28: Late steering Input 2) Over-speeding is the main reason for crash among experts. Black lines representing the speed of the vehicle in Figure 5 shows the same. Figure 29: Over-speeding 70 Appendix 4: TLX Questionnaire The questions below are about your experience in the session that you just performed. Put a cross on the line, not between them. Mental Demand Very Low How mentally demanding was the task? 0 Very High 20 Physical Demand Very Low How physically demanding was the task? 0 Very High 20 Performances Perfect How successful were you in accomplishing what you were asked to do? 0 Failure 20 Effort Very Low How hard did you have to work to accomplish your level of performance? 0 Very High 20 Frustrations Very Low How insecure, discouraged, irritated, stressed, and annoyed were you? 0 Very High 20 71 DOUBLE LANE CHANGE MENTAL PHYSICAL PERFORMANCE EFFORT FRUSTATION DRIVER 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 3 6 13 11 13 10 15 4 12 3 13 14 10 7 12 6 6 14 7 5 15 14 15 9 8 7 9 10 14 13 13 9 13 8 9 10 9 2 7 3 8 3 1 13 4 5 1 6 9 11 12 12 6 9 9 7 1 11 4 12 8 4 3 7 6 8 9 7 5 3 8 6 10 4 3 12 14 7 3 5 4 5 13 2 9 9 13 10 13 11 15 7 10 16 10 10 12 4 9 7 14 12 11 8 9 11 8 12 7 13 7 9 15 16 10 5 5 14 12 13 6 11 12 13 14 7 7 11 14 8 10 14 13 12 13 12 9 12 12 12 13 11 12 10 10 9 11 12 1 16 13 6 14 2 15 2 12 1 2 3 8 3 11 10 1 13 8 4 15 14 10 8 7 6 6 7 12 13 2 8 13 9 5 5 8 72 HIGH SPEED CORNERING SESSION 1 MENTAL PHYSICAL PERFORMANCE EFFORT FRUSTATION DRIVER 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 10 18 16 15 13 16 17 8 17 13 15 17 14 15 11 10 16 19 15 7 15 14 15 12 12 15 16 15 14 17 16 13 14 12 14 15 11 7 17 17 16 14 12 16 6 12 8 11 12 11 15 11 5 14 17 14 11 10 4 12 11 9 10 13 9 8 14 11 14 15 8 10 13 5 10 12 11 14 6 17 16 12 18 8 14 11 11 7 11 9 9 17 12 6 11 10 12 14 15 13 13 9 11 17 12 11 8 10 10 10 11 12 16 16 4 14 17 17 15 15 10 16 16 14 14 9 12 13 18 16 14 13 13 12 16 15 12 14 13 12 16 16 16 15 13 11 12 14 9 12 12 4 12 14 17 11 8 8 8 5 10 5 10 8 11 16 12 6 10 14 10 14 9 8 9 12 12 12 6 11 12 9 9 10 12 73 HIGH SPEED CORNERING SESSION 2 MENTAL PHYSICAL PERFORMANCE EFFORT FRUSTATION DRIVER 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 7 18 1 12 7 10 11 3 17 11 9 16 10 10 13 11 14 18 12 15 16 14 15 7 10 14 15 14 14 14 12 16 4 10 13 14 8 6 14 2 14 9 6 12 7 15 6 9 10 10 15 14 5 16 13 17 10 12 4 12 9 11 10 15 10 8 13 9 11 5 7 9 14 5 11 9 12 6 8 13 10 9 13 13 7 13 12 15 10 7 7 12 3 8 8 10 12 9 11 13 10 8 11 11 10 10 10 9 12 9 9 15 16 4 12 14 10 7 16 16 14 14 11 12 15 14 10 16 14 8 12 14 13 12 11 16 15 16 14 12 11 12 13 9 11 13 14 14 9 17 3 13 10 13 7 6 12 13 5 10 9 10 13 9 3 17 13 9 15 14 10 6 9 12 7 8 12 9 7 13 6 9 11 6 10 74 Appendix 5: Vehicle dynamics questionnaire Table: Vehicle Dynamics Question 1 Question: What will happen if the driver Experienced Novices Total A. Vehicle safely comes to a stop 0% 0% 0% B. Rear Wheels start to skid 86% 73% 77% C. Front Wheels start to skid 14% 27% 23% suddenly brakes while taking a corner? This driving situation is often referred to as brake in over-steer. When the driver suddenly brakes, weight of the vehicle is transferred to the front wheels. Thus the rear wheels, which now carry less weight, provide lesser lateral force. This means that the vehicle will have an over-steer tendency and hence the rear will try to skid out. Table: Vehicle Dynamics Question 2 Question: While taking a right hand corner the driver notices that rear of the vehicle is Experienced Novices Total A. Brake 0% 0% 0% B. Increase Throttle 0% 6.67% 5% C. Steer into the corner 29% 6.67% 14% D. Steer out of the corner 71% 86.67 % 82% skidding out. What action should he take? This driving situation is encountered while experiencing low friction surfaces unexpectedly and is referred to as counter-steering in which the driver steers the vehicle out of the turn when the rear is skidding in order to keep the vehicle on track. As can be seen from Figure 12 below 82% participants answered that the driver should steer out of the corner. 75 Table: Vehicle Dynamics Question 3 Question: The driver suddenly brakes on a straight road with left side of the car on water. Experienced Novices Total A. Vehicle stops in straight position 14% 6.67% 9% B. Vehicle steers to the right 43% 26.67% 32% C. Vehicle steers to the left 43% 66.67% 59% What will happen? This driving situation is often referred to as split-mu braking. When one side of the vehicle is on water (in this case left side) and the driver brakes then the brake force generated on the water side is lower because of lower friction of coefficient. Thus the unbalanced moment caused by different brake forces on the two sides of the vehicle will tend to make the vehicle go away from the low friction surface i.e. towards the right. 76 velx(1:len)=filter(hd1,S.RFLocalVelX); Appendix 6: Matlab Programs 1. Race-Track experiment %% STEERING PERFORMANCE close all; clear all; last=0; %% DEFINING FILTERS [z,p,k]=butter(2,3/50); %% 3HZ LOW PASS FILTER [sos,g]=zp2sos(z,p,k); hd=dfilt.df2sos(sos,g); [z1,p1,k1]=butter(2,10/50); %% 10HZ LOW PASS FILTER [sos1,g1]=zp2sos(z1,p1,k1); hd1=dfilt.df2sos(sos1,g1); [zl,pl,kl]=butter(2,2/50); %% 2HZ LOW PASS FILTER [sosl,gl]=zp2sos(zl,pl,kl); hdl=dfilt.df2sos(sosl,gl); [zh,ph,kh]=butter(2,0.5/50,'high'); %% 0.5HZ HIGH PASS FILTER [sosh,gh]=zp2sos(zh,ph,kh); hdh=dfilt.df2sos(sosh,gh); for curve=1:3 for pp=1:17 clearvars -except curve pp Steer last hd hd1 hdh hdl crash_num C1=0; C2=0; C3=0; tot_crash=0; tot_dep=0; num=0; x1_origin=350; y1_origin=175; x2_origin=0; y2_origin=100; x3_origin=-180; y3_origin=-180; w=0; e=0; Fs=100; session=4; % % Take fastest lap reference point x_curve=[50 192 -150]; z_curve=[447 -80 193]; S=load(sprintf('%s%d%s%d','PP_',pp,'_session_',session)); len=length(S.RFPosX); x(1:len)=filter(hd1,S.RFPosX); y(1:len)=filter(hd1,S.RFPosY); z(1:len)=filter(hd1,S.RFPosZ); vely(1:len)=filter(hd1,S.RFLocalVelY); velz(1:len)=filter(hd1,S.RFLocalVelZ); accx(1:len)=filter(hd1,S.RFLocalAccelX); accy(1:len)=filter(hd1,S.RFLocalAccelY); accz(1:len)=filter(hd1,S.RFLocalAccelZ); Lap(1:len)=[S.RFLapNumber]; Laptime(1:len)=[S.LogTimeFromStart]; throttle(1:len)=filter(hd1,S.RFUnfilteredThrottle); brake(1:len)=filter(hd1,S.RFUnfilteredBrake); steering(1:len)=100*[S.RFUnfilteredSteering]; front_tire_load(1:len)=S.RFFrontLeftTireLoad+S.RFFrontR ightTireLoad; rear_tire_load(1:len)=S.RFRearLeftTireLoad+S.RFRearRig htTireLoad; SurfFR(1:len)=S.RFFrontRightSurfaceType; SurfFL(1:len)=S.RFFrontLeftSurfaceType; SurfRR(1:len)=S.RFRearRightSurfaceType; SurfRL(1:len)=S.RFRearLeftSurfaceType; Grip(1:len)=S.RFFrontLeftGripFract; steer_fit(1:len)=filter(hd,steering(1:len)); %% APPLYING THE 3HZ LOW PASS FILTER ON THE STEERING ANGLE steer_diff(1:len-1)=diff(steer_fit(1:len)); %% CALCULATING THE STEERING WHEEL RATE BY DIFFERENTIATING THE STEERING ANGLE SIGNAL steer_jerk(1:len-3)=abs(diff(steer_fit(1:len),3)); %% CALCULATING THE STEERING WHEEL JERK BY TRIPLE DIFFERENTIATION OF THE STEERING ANGLE SIGNAL steerlow(1:len)=filter(hdl,steering(1:len)); %% APPLYING THE 0.6HZ LOW PASS FILTER ON THE STEERING ANGLE steerhigh(1:len)=filter(hdh,steerlow(1:len)); %% APPLYING THE 0.3HZ HIGH PASS FILTER ON THE STEERING ANGLE %% LAP SELECTION l=Lap(1); lap(1)=1; k=2; for i=1:len if round(Lap(i))~=l if Lap(i)==0 lap(k)=i-1; l=round(Lap(i+1)); lap(k)=i+1; else lap(k)=i; l=round(Lap(i)); end k=k+1; end end lap(k)=len; % for i=1:size(lap,2)-1 % if lap(i+1)>0 % lap_time(i)=Laptime(lap(i+1))-Laptime(lap(i)); % end % end % k=2; % nextlap = input('Would you like to analyze a lap', 's'); % while strcmp(nextlap, 'Y') && (k<=size(lap,2)-1) % speed=3.6*sqrt(velx(lap(k-1):lap(k)).^2+vely(lap(k1):lap(k)).^2+velz(lap(k-1):lap(k)).^2); % % figure(1) % subplot(4,1,1) % plot(speed) % title('speed in km/h') % % subplot(4,1,2) % plot(throttle(lap(k-1):lap(k))); % title('Throttle') % % subplot(4,1,3) % plot(brake(lap(k-1):lap(k))); % title('Brake') % % subplot(4,1,4) % plot(steering(lap(k-1):lap(k))); % title('Steering') % % figure(2) % subplot(3,1,1) % plot(z(lap(k-1):lap(k))) % % subplot(3,1,2) % plot(x(lap(k-1):lap(k))) % % subplot(3,1,3) % plot(z(lap(k-1):lap(k)),-x(lap(k-1):lap(k))) % % % % k=k+1; % nextlap = input('Would you like to analyze the next lap', 's'); % % if (nextlap = 'Y') % end % % if strcmp(nextlap, 'N') || (k>size(lap,2)-1) % fprintf('\nBye!\n'); % % end %% CURVE SELECTION for i=1:length(lap)-1 flag=1; for j=lap(i):lap(i+1) if j>0 if z(j)>255 && flag==1 entry_lap(i,1)=j; k=j; while(z(k)>255&&k-150&&k-140 exit_lap(i,3)=k; else exit_lap(i,3)=0; entry_lap(i,3)=0; end flag=4; j=k; end %% CURVE 1 %% CURVE 2 %% CURVE 3 end 78 end end %% CRASHES for i=1:length(lap)-1 cha1=1; cha2=1; cha3=1; if lap(i)>0&&lap(i+1)>0 for j=lap(i):lap(i+1) speed=3.6*sqrt(velx(j).^2+vely(j).^2+velz(j).^2); % LOSS OF CONTROL CONDITION: NONE OF THE WHEELS ARE ON ASPHALT if SurfFR(j)~=0&&SurfFL(j)~=0&&SurfRR(j)~=0&&SurfRL (j)~=0 if j0 if j>entry_lap(i,2)&&j0 if j>entry_lap(i,3)&&j0 num=num+1; end if crash(j)>111 tot_crash=tot_crash+1; if crash(j)>200 C1=C1+1; tot_dep=tot_dep+1; end if mod(crash(j),100)>20 C2=C2+1; tot_dep=tot_dep+1; end if mod(crash(j),10)>1 C3=C3+1; tot_dep=tot_dep+1; end end crash_num(pp,1)=C1; crash_num(pp,2)=C2; crash_num(pp,3)=C3; crash_num(pp,4)=num; crash_num(pp,5)=tot_crash; crash_num(pp,6)=tot_dep; end %% CURVE STEERING PERFORMANCE ANALYSIS for i=2:size(entry_lap,1) prod=0; count=0; k=entry_lap(i,curve); if entry_lap(i,curve)~=0 steer_max=max(abs(steer_fit(entry_lap(i,curve):exit_lap(i,c urve)))); brake_max=max(brake(entry_lap(i,curve):exit_lap(i,curve))) ; n=exit_lap(i,curve)-entry_lap(i,curve); %% Number of samples time=(1:n)/100; T=n/100; XFs=fft(steerlow(entry_lap(i,curve):exit_lap(i,curve)))/Fs; XFshigh=fft(steerhigh(entry_lap(i,curve):exit_lap(i,curve)))/ Fs; % fftshift moves this to [-0.5, 0.5] for better visualization Xf=fftshift(XFs); Xfhigh=fftshift(XFshigh); % Computing energy spectrum Exf=abs(Xf).^2; Exfhigh=abs(Xfhigh).^2; % Pxf is an approximation to the continuous time PSD Pxf=Exf/T; Pxfhigh=Exfhigh/T; % Assigning frequencies to the samples of the PSD % figure(2) df=Fs/n; % Df=freq separation between two consecutive fft points 79 freq=[-(n/2)+1:1:n/2+1]*df; % maxPxf=max(Pxf); % subplot(211) % axis([-50 50 0 maxPxf]) % title('Power Spectral Density of x(t) '); % subplot(212) % plot(freq,10*log10(Pxflow/maxPxf)) % axis([-50 50 -100 0]) % grid % title('Power Spectral Density of x(t) in dB '); Power_from_PSD=sum(Pxf)*df; Power_from_PSDhigh=sum(Pxfhigh)*df; if entry_lap(i,curve)~=0 && i<=size(pos3,1) && pos3(i,1)~=0 sum1=sum(abs(steer_diff(pos3(i,1):exit_lap(i,curve)))); sum2=sum(abs(steer_jerk(pos3(i,1):exit_lap(i,curve)))); max1=max(steer_jerk(pos3(i,1):exit_lap(i,curve))); avg_steer_speed(i,curve)=sum1/(0.01*(pos3(i,1)+exit_lap(i,curve))); avg_steer_jerk(i,curve)=sum2/(0.000001*(pos3(i,1)+exit_lap(i,curve))); max_steer_jerk(i,curve)=max1/0.000001; end end HFC(i,curve)=Power_from_PSDhigh/Power_from_PSD; %% HIGH FREQUENCY COMPONENT OF THE STEERING ANGLE %% STEERING REVERSAL RATE (3 DEGREE GAP) for j=entry_lap(i,curve):exit_lap(i,curve)-2 prod=prod+brake(j)*abs(steer_fit(j))/(steer_max*brake_max ); if sign(steer_diff(j))~=sign(steer_diff(j+1))&&steer_diff(j)~=0 &&steer_diff(j+1)~=0&&abs(steer_fit(j)-steer_fit(k))>3 flag =0; p=j+1; while(flag==0&&p3 count=count+1; k=j+1; pos3(i,count)=j+1; end else p=p+1; end end end end end steer_brake(i,curve)=prod/((entry_lap(i,curve)+exit_lap(i,curve))*.01); if count>0 SRR(i,curve)=count/((pos3(i,count)-pos3(i,1))*.01); else SRR(i,curve)=count; end %% STEERING WHEEL RATE AND STEERING JERK CALCULATION %% OTHER MEASURES for i=2:size(entry_lap,1) if entry_lap(i,curve)~=0 j=entry_lap(i,curve); while (brake(j)==0&&j0 for j=entry_lap(i,curve):exit_lap(i,curve)-1 k=j-entry_lap(i,curve)+1; d=d+sqrt((z(j+1)-z(j)).^2+(x(j+1)-x(j)).^2); X_diff=x3_origin-z(j); Y_diff=y3_origin-x(j); R1(i,k)=sqrt(X_diff.^2+Y_diff.^2); Theta(i,k)=atan(Y_diff/X_diff); V=sqrt(velx(j).^2+vely(j).^2+velz(j).^2); ACC=sqrt(accx(j).^2+accy(j).^2+accz(j).^2); 80 TR3(i,k)=V.^2/ACC; CURV3(i,k)=ACC/V.^2; end dis(i,curve)=d; SORT1=sort(CURV3(i,:),2,'descend'); SORT1=SORT1(:,1:5,:); SUM(i,curve)=sum(SORT1,2); end end %% WRINTING INTO A SINGLE STEERING ARRAY n=last+(curve-1)*size(entry_lap,2)+1; for p=n:n+size(entry_lap,1)-2 Steer(p,1)=pp; Steer(p,2)=curve; Steer(p,3)=SRR(p+1-n,curve); Steer(p,4)=avg_steer_speed(p+1-n,curve); Steer(p,5)=brake_point_x(p+1-n,curve); Steer(p,6)=brake_point_z(p+1-n,curve); Steer(p,7)=curve_time(p+1-n,curve); Steer(p,8)=HFC(p+1-n,curve); Steer(p,9)=dis(p+1-n,curve); Steer(p,10)=SUM(p+1-n,curve); Steer(p,11)=avg_steer_jerk(p+1-n,curve); Steer(p,12)=max_steer_jerk(p+1-n,curve); Steer(p,13)=gglimit(p+1-n,curve); Steer(p,14)=g_util(p+1-n,curve); Steer(p,15)=brake_dist(p+1-n,curve); Steer(p,16)=steer_brake(p+1-n,curve); Steer(p,17)=crash(p+1-n); end last=p; end end xlswrite('Steering.xls',Steer); %% SEPERATION OF THE DATA INTO CRASH (LOSS OF CONTROL) AND NO-CRASH DATA NC=0; C=0; for j=1:size(Steer,1) if Steer(j,3)>0&&Steer(j,4)>0 if Steer(j,17)==111 NC=NC+1; STEER_NC(NC,:)=Steer(j,:); %% NO CRASH DATA else C=C+1; STEER_C(C,:)=Steer(j,:); %% CRASH DATA end end end ex1=0; nex1=0; ex2=0; nex2=0; ex3=0; nex3=0; for i=1:size(STEER_NC,1) if STEER_NC(i,1)==1||STEER_NC(i,1)==11||STEER_NC(i,1 )==12||STEER_NC(i,1)==13||STEER_NC(i,1)==14||STEER _NC(i,1)==16||STEER_NC(i,1)==17 if STEER_NC(i,2)==1 ex1=ex1+1; experts_curve1(ex1,:)=STEER_NC(i,:); end if STEER_NC(i,2)==2 ex2=ex2+1; experts_curve2(ex2,:)=STEER_NC(i,:); end if STEER_NC(i,2)==3 ex3=ex3+1; experts_curve3(ex3,:)=STEER_NC(i,:); end else if STEER_NC(i,2)==1 nex1=nex1+1; nonexperts_curve1(nex1,:)=STEER_NC(i,:); end if STEER_NC(i,2)==2 nex2=nex2+1; nonexperts_curve2(nex2,:)=STEER_NC(i,:); end if STEER_NC(i,2)==3 nex3=nex3+1; nonexperts_curve3(nex3,:)=STEER_NC(i,:); end end end %% CALCULATING THE MEAN OF THE CALCULATED MEASURES FOR NO-CRASH DATA count_start=1; count_end=0; for curve=1:3 count1=0; count2=0; for pp=1:17 s1=0; count=0; for i=1:size(STEER_NC,1) if STEER_NC(i,1)==pp && STEER_NC(i,2)==curve s1=s1+STEER_NC(i,:); count=count+1; count_end=count_end+1; 81 if STEER_NC(i,1)==1||STEER_NC(i,1)==11||STEER_NC(i,1 )==12||STEER_NC(i,1)==13||STEER_NC(i,1)==14||STEER _NC(i,1)==16||STEER_NC(i,1)==17 count1=count1+1; temp1(count1,curve)=STEER_NC(i,9); else count2=count2+1; temp2(count2,curve)=STEER_NC(i,9); end end end MEAN(pp,1:size(STEER_NC,2),curve)=s1/count; MEAN1(pp,1:size(STEER_NC,2),curve)=mean(STEER_N C(count_start:count_end,:),1); stdev(pp,1,curve)=pp; stdev(pp,2,curve)=curve; if count_start~=count_end stdev(pp,3:size(STEER_NC,2),curve)=std(STEER_NC(cou nt_start:count_end,3:17),1); else stdev(pp,1:size(STEER_NC,2),curve)=zeros(1,17); end count_start=count_end+1; end end for i=1:3 k=1; while (temp1(k,i)>0 && k0&& p0 if DEV(i,1)==1||DEV(i,1)==11||DEV(i,1)==12||DEV(i,1)==13| |DEV(i,1)==14||DEV(i,1)==16||DEV(i,1)==17 if DEV(i,2)==1 c1=c1+1; EXPERT_Curve1_dev(c1,:)=DEV(i,:); end if DEV(i,2)==2 c2=c2+1; EXPERT_Curve2_dev(c2,:)=DEV(i,:); end if DEV(i,2)==3 c3=c3+1; EXPERT_Curve3_dev(c3,:)=DEV(i,:); end else if DEV(i,2)==1 nc1=nc1+1; NON_EXPERT_Curve1_dev(nc1,:)=DEV(i,:); end if DEV(i,2)==2 82 nc2=nc2+1; NON_EXPERT_Curve2_dev(nc2,:)=DEV(i,:); end if DEV(i,2)==3 nc3=nc3+1; NON_EXPERT_Curve3_dev(nc3,:)=DEV(i,:); end end end end %% PERFORMING T-TEST [h1,p1]=ttest2(EXPERT_Curve1(:,3:16),NON_EXPERT_C urve1(:,3:16),[],[],'unequal'); [h2,p2]=ttest2(EXPERT_Curve2(:,3:16),NON_EXPERT_C urve2(:,3:16),[],[],'unequal'); [h3,p3]=ttest2(EXPERT_Curve3(:,3:16),NON_EXPERT_C urve3(:,3:16),[],[],'unequal'); %% PERFORMING WILCOXON RANK SUM TEST for i=3:16 [pw1(i-2),hw1(i2)]=ranksum(EXPERT_Curve1(:,i),NON_EXPERT_Curve1 (:,i)); [pw2(i-2),hw2(i2)]=ranksum(EXPERT_Curve2(:,i),NON_EXPERT_Curve2 (:,i)); [pw3(i-2),hw3(i2)]=ranksum(EXPERT_Curve3(:,i),NON_EXPERT_Curve3 (:,i)); [pw1(i-2),hw1(i2)]=ranksum(EXPERT_Curve1(:,i),NON_EXPERT_Curve1 (:,i)); [pw2(i-2),hw2(i2)]=ranksum(EXPERT_Curve2(:,i),NON_EXPERT_Curve2 (:,i)); [pw3(i-2),hw3(i2)]=ranksum(EXPERT_Curve3(:,i),NON_EXPERT_Curve3 (:,i)); end FINAL1(1:size(EXPERT_Curve1,1),:)=EXPERT_Curve1; FINAL1(size(EXPERT_Curve1,1)+1,:)=mean(EXPERT_Cu rve1); FINAL1(size(EXPERT_Curve1,1)+2,:)=std(EXPERT_Curv e1); FINAL1(size(EXPERT_Curve1,1)+4:size(EXPERT_Curve1 ,1)+3+size(NON_EXPERT_Curve1,1),:)=NON_EXPERT_ Curve1; FINAL1(21,:)=mean(NON_EXPERT_Curve1); FINAL1(22,:)=std(NON_EXPERT_Curve1); FINAL1(23,3:16)=pw1; % FINAL1(24,3:16)=pw1; FINAL2(1:size(EXPERT_Curve2,1),:)=EXPERT_Curve2; FINAL2(size(EXPERT_Curve2,1)+1,:)=mean(EXPERT_Cu rve2); FINAL2(size(EXPERT_Curve2,1)+2,:)=std(EXPERT_Curv e2); FINAL2(size(EXPERT_Curve2,1)+4:size(EXPERT_Curve2 ,1)+3+size(NON_EXPERT_Curve2,1),:)=NON_EXPERT_ Curve2; FINAL2(21,:)=mean(NON_EXPERT_Curve2); FINAL2(22,:)=std(NON_EXPERT_Curve2); FINAL2(23,3:16)=pw2; % FINAL2(24,3:16)=pw2; FINAL3(1:size(EXPERT_Curve3,1),:)=EXPERT_Curve3; FINAL3(size(EXPERT_Curve3,1)+1,:)=mean(EXPERT_Cu rve3); FINAL3(size(EXPERT_Curve3,1)+2,:)=std(EXPERT_Curv e3); FINAL3(size(EXPERT_Curve3,1)+4:size(EXPERT_Curve3 ,1)+3+size(NON_EXPERT_Curve3,1),:)=NON_EXPERT_ Curve3; FINAL3(21,:)=mean(NON_EXPERT_Curve3); FINAL3(22,:)=std(NON_EXPERT_Curve3); FINAL3(23,3:16)=pw3; % FINAL3(24,3:16)=pw3; %% DATA WRINTING INTO EXCEL xlswrite('CURVE1.xls',FINAL1); xlswrite('CURVE2.xls',FINAL2); xlswrite('CURVE3.xls',FINAL3); %% SEPERATION OF THE NO-CRASH DATA INTO EXPERTS AND NON-EXPERTS FOR EACH CURVE (ALL DATA) t1=0; r1=0; t2=0; r2=0; t3=0; r3=0; for j=1:size(STEER_NC,1) if STEER_NC(j,3)>0&&STEER_NC(j,4)>0&&STEER_NC(j, 5)>0&&STEER_NC(j,6)>0 if STEER_NC(j,1)==1||STEER_NC(j,1)==11||STEER_NC(j,1 )==12||STEER_NC(j,1)==13||STEER_NC(j,1)==14||STEER _NC(j,1)==16||STEER_NC(i,1)==17 if STEER_NC(j,2)==1 t1=t1+1; Steer_NC_expert1(t1,:)=STEER_NC(j,:); %% EXPERTS NO-CRASH CURVE 1 end if STEER_NC(j,2)==2 t2=t2+1; Steer_NC_expert2(t2,:)=STEER_NC(j,:); %% EXPERTS NO-CRASH CURVE 2 end if STEER_NC(j,2)==3 83 t3=t3+1; Steer_NC_expert3(t3,:)=STEER_NC(j,:); %% EXPERTS NO-CRASH CURVE 3 end else if STEER_NC(j,2)==1 r1=r1+1; Steer_NC_nonexpert1(r1,:)=STEER_NC(j,:); %% NON-EXPERTS NO-CRASH CURVE 1 end if STEER_NC(j,2)==2 r2=r2+1; Steer_NC_nonexpert2(r2,:)=STEER_NC(j,:); %% NON-EXPERTS NO-CRASH CURVE 2 end if STEER_NC(j,2)==3 r3=r3+1; Steer_NC_nonexpert3(r3,:)=STEER_NC(j,:); %% NON-EXPERTS NO-CRASH CURVE 3 end end end end %% SEPERATION OF THE CRASH DATA INTO EXPERTS AND NON-EXPERTS FOR EACH CURVE (ALL DATA) t1=0; r1=0; t2=0; r2=0; t3=0; r3=0; for j=1:size(STEER_C,1) if STEER_C(j,3)>0&&STEER_C(j,4)>0&&STEER_C(j,5)>0 &&STEER_C(j,6)>0 if STEER_C(j,1)==1||STEER_C(j,1)==11||STEER_C(j,1)==12 ||STEER_C(j,1)==13||STEER_C(j,1)==14||STEER_C(j,1)== 16||STEER_C(i,1)==17 else if STEER_C(j,2)==1 r1=r1+1; Steer_C_nonexpert1(r1,:)=STEER_C(j,:); %% NON-EXPERTS CRASH CURVE 1 end if STEER_C(j,2)==2 r2=r2+1; Steer_C_nonexpert2(r2,:)=STEER_C(j,:); %% NON-EXPERTS CRASH CURVE 2 end if STEER_C(j,2)==3 r3=r3+1; Steer_C_nonexpert3(r3,:)=STEER_C(j,:); %% NON-EXPERTS CRASH CURVE 3 end end end end %% NUMBER OF CRASHES crash_num(18,1)=sum(crash_num(:,1)); crash_num(18,2)=sum(crash_num(:,2)); crash_num(18,3)=sum(crash_num(:,3)); crash_num(18,4)=sum(crash_num(:,4)); crash_num(18,5)=sum(crash_num(:,5)); crash_num(18,6)=sum(crash_num(:,6)); e=0; n=0; for k=1:size(crash_num,1)-1 if k==1||k==11||k==12||k==13||k==14||k==16||k==17 e=e+1; crash_exp(e,:)=crash_num(k,:); else n=n+1; crash_nexp(n,:)=crash_num(k,:); end end if STEER_C(j,2)==1 t1=t1+1; Steer_C_expert1(t1,:)=STEER_C(j,:); %% EXPERTS CRASH CURVE 1 end if STEER_C(j,2)==2 t2=t2+1; Steer_C_expert2(t2,:)=STEER_C(j,:); %% EXPERTS CRASH CURVE 2 end if STEER_C(j,2)==3 t3=t3+1; Steer_C_expert3(t3,:)=STEER_C(j,:); %% EXPERTS CRASH CURVE 3 end 84 2. Double Lane Change test hd1=dfilt.df2sos(sos1,g1); %% DOUBLE LANE CHANGE TEST clear all close all y_final_expert(1:10,1:131)=0; y_final_nonexpert(1:10,1:131)=0; hit(1:37,1:8)=0; for D=1:37 %% LOADING DRIVER DATA [zl,pl,kl]=butter(2,2/50); %% 2HZ LOW PASS FILTER [sosl,gl]=zp2sos(zl,pl,kl); hdl=dfilt.df2sos(sosl,gl); P=load(sprintf('%s%d%s','Driver_',D,'_Lane_Change')); for i=1:40 if i<10 data(i)=P.(sprintf('%s%d%s%d','driver_',D,'_trial_00',i)); else data(i)=P.(sprintf('%s%d%s%d','driver_',D,'_trial_0',i)); end y_final=0; set(gcf,'Color',[1,1,1]); end for session=1:8 clearvars -except session RMS rms_accy data yi pos X_pos M rate jerk max_steer RMS_MID D STEER y_final_expert y_final_nonexpert hit avg_steer_rate avg_steer_jerk rate_c rate_comp jerk_c jerk_comp %% Pylon Positions if session==1 || session==8 X1=[150 153.75 157.5 161.25 165]; X2=[195 201.25 207.5 213.75 220]; X3=[245 248.75 252.5 256.25 260]; Y1=[3 3 3 3 3]; Y2=[0 0 0 0 0 ]; else X1=[350 353.75 357.5 361.25 365]; X2=[395 401.25 407.5 413.75 420]; X3=[445 448.75 452.5 456.25 460]; Y1=[3 3 3 3 3]; Y2=[0 0 0 0 0 ]; end [zh,ph,kh]=butter(2,0.5/50,'high'); %% 0.5HZ HIGH PASS FILTER [sosh,gh]=zp2sos(zh,ph,kh); hdh=dfilt.df2sos(sosh,gh); %% CALCULATING THE RMS ACCELERATION for i=(session-1)*5+1:session*5 d=i-(session-1)*5; steer=filter(hd,data(i).Y(1,13).Data); steer_jerk=diff(steer,3); steer_diff=diff(steer); accy=downsample(data(i).Y(1,6).Data,5)/9.81; count=1; if session==1 || session==8 while data(i).Y(1,23).Data(count)<150 && count 150 meters count=count+1; end t_start=count; while data(i).Y(1,23).Data(count)<270 && count 350 meters count=count+1; end t_start=count; start=[900 1750 1750 1700 1650 1650 1600 900]; stop=[1650 2400 2400 2300 2250 2200 2150 1650]; if session==1 || session==8 in=[140 180 230]; out=[180 230 260]; else in=[340 380 430]; out=[380 430 460]; end %% DEFINING FILTERS [z,p,k]=butter(2,3/50); %% 3HZ LOW PASS FILTER [sos,g]=zp2sos(z,p,k); hd=dfilt.df2sos(sos,g); while data(i).Y(1,23).Data(count)<470 && count10 && c<=15 && yi(d,cones(c))+1>=3 hit(D,session)=hit(D,session)+1; end if c>10 && c<=15 && yi(d,cones(c))-1<0 hit(D,session)=hit(D,session)+1; end end y_final=y_final+yi(d,:); for w=1:3 p=find(data(i).Y(1,23).Data>in(w)); q=find(data(i).Y(1,23).Data>out(w)); if w==2 [M,I]=max((steer(p(1):q(1)))); else [M,I]=min((steer(p(1):q(1)))); end % grid on % hold on % plot(X1,Y1,'--r'); % plot(X1,Y2,'--r'); % plot(X2,-Y1,'--r'); % plot(X2,Y2,'--r'); % plot(X3,Y1,'--r'); % plot(X3,Y2,'--r'); %% % plot((data(i).Y(1,23).Data(1,start(session):stop(session))+dat a(i).Y(1,26).Data(1,start(session):stop(session)))/2,(data(i).Y (1,24).Data(1,start(session):stop(session))+data(i).Y(1,27).D ata(1,start(session):stop(session)))/2) xx=(data(i).Y(1,23).Data(1,start(session):stop(session))+dat a(i).Y(1,26).Data(1,start(session):stop(session)))/2; yy=(data(i).Y(1,24).Data(1,start(session):stop(session))+dat a(i).Y(1,27).Data(1,start(session):stop(session)))/2; yi(d,:)=interp1(xx,yy,xi); %% Number of cones hit cones=[10 14 18 21 25 55 61 67 74 80 105 109 113 116 120]; for c=1:15 if c<=5 && yi(d,cones(c))+1>=3 hit(D,session)=hit(D,session)+1; end if c<=5 && yi(d,cones(c))-1<0 hit(D,session)=hit(D,session)+1; end if c>5 && c<=10 && yi(d,cones(c))+1>=0 hit(D,session)=hit(D,session)+1; end if c>5 && c<=10 && yi(d,cones(c))-1<-3 hit(D,session)=hit(D,session)+1; max_steer(d,w,session)=abs(M); pos(d,w+1,session)=p(1)+I; X_pos(d,w+1,session)=(data(i).Y(1,23).Data(1,p(1)+I)+data (i).Y(1,26).Data(1,p(1)+I))/2; end pos(d,1,session)=pos(d,2,session)-100; pos(d,5,session)=pos(d,4,session)+100; % figure(2) % grid on % subplot(8,1,session) % axis([xi(1) xi(length(xi)) -100 100]) % hold on % plot((data(i).Y(1,23).Data(1,start(session):stop(session))+dat a(i).Y(1,26).Data(1,start(session):stop(session)))/2,steer(start (session):stop(session))); %% PLOTTING THE RMS ACCELERATION % figure(3) % grid on % subplot(8,1,session) % axis([xi(1) xi(length(xi)) -1 1]) % hold on % plot((data(i).Y(1,23).Data(1,start(session):stop(session))+dat a(i).Y(1,26).Data(1,start(session):stop(session)))/2,accy(start (session):stop(session))); end y_final=y_final/5; if D==1|| D==7 || D==8 || D==9 || D==10 || D==17 || D==22 || D==23|| D==24 || D==25|| D==26 || D==27|| D==28 || D==29 || D==30|| D==35||D==36||D==37 86 y_final_expert(session,5:131)=y_final_expert(session,5:131) +y_final(1,5:length(y_final)); steer_jerk=diff(steer,3); steer_diff=diff(steer); for n=1:4 else y_final_nonexpert(session,5:131)=y_final_nonexpert(sessio n,5:131)+y_final(1,5:length(y_final)); end rms_accy(6,session)=mean(rms_accy(1:5,session)); rms_accy(7,session)=std(rms_accy(1:5,session)); %% CALCUTING THE RMS DEVIATION FROM THE MEAN PATH for g=1:5 sum1=0; for i=1:size(yi,2) if isnan(y_final(1,i))~=1 sum1=sum1+(y_final(1,i)-yi(g,i))^2; end end sum_mean=sum((yi(g,10:25)1.5).^2)+sum((yi(g,55:80)+1.5).^2)+sum((yi(g,105:120)1.5).^2); RMS_MID(g,session)=sqrt(sum_mean/55); RMS(g,session)=sqrt(sum1/131); end SD=0; S_mid=0; S_mean=0; for sel=1:5 if RMS_MID(sel,session)<1.5 SD=SD+1; S_mid=S_mid+RMS_MID(sel,session); S_mean=S_mean+RMS(sel,session); end end RMS(6,session)=S_mean/SD; RMS_MID(6,session)=S_mid/SD; if D==4 || D==8 ||D==9 || D==11 % figure(1) % subplot(8,1,session) % axis([xi(1) xi(length(xi)) -5 5]) % hold on % plot(xi,y_final,'r') % else % figure(2) % subplot(8,1,session) % axis([xi(1) xi(length(xi)) -5 5]) % hold on % plot(xi,y_final,'g') end for u=(session-1)*5+1:session*5 d=u-(session-1)*5; % rate=0; % jerk=0; steer=filter(hd,data(u).Y(1,13).Data); rate(n,d,session)=sum(abs(steer_diff(pos(d,n,session):pos(d, n+1,session))))/(0.01*(pos(d,n+1,session)pos(d,n,session))); jerk(n,d,session)=sum(abs(steer_jerk(pos(d,n,session):pos(d, n+1,session))))/(0.000001*(pos(d,n+1,session)pos(d,n,session))); end rate_c(d,session)=sum(abs(steer_diff(pos(d,3,session):pos(d, 5,session))))/(0.01*(pos(d,5,session)-pos(d,3,session))); jerk_c(d,session)=sum(abs(steer_jerk(pos(d,3,session):pos(d ,5,session))))/(0.000001*(pos(d,5,session)pos(d,3,session))); end rate_comp(session)=mean(rate_c(1:5,session)); jerk_comp(session)=mean(jerk_c(1:5,session)); for n=1:4 rate(n,6,session)=mean(rate(n,1:5,session)); jerk(n,6,session)=mean(jerk(n,1:5,session)); rate(n,7,session)=std(rate(n,1:5,session)); jerk(n,7,session)=std(jerk(n,1:5,session)); end avg_steer_rate(session)=mean(rate(1:4,6,session)); avg_steer_jerk(session)=mean(jerk(1:4,6,session)); end for session=1:8 STEER(D,1,session)=rms_accy(7,session); STEER(D,2,session)=RMS(6,session); STEER(D,3,session)=RMS_MID(6,session); STEER(D,4,session)=mean(X_pos(1:5,2,session)); STEER(D,5,session)=mean(X_pos(1:5,3,session)); STEER(D,6,session)=mean(X_pos(1:5,4,session)); STEER(D,7,session)=rate_comp(session); STEER(D,8,session)=jerk_comp(session); STEER(D,9,session)=mean(max_steer(1:5,3,session)); STEER(D,10,session)=hit(D,session)/5; STEER(D,11,session)=avg_steer_rate(session); STEER(D,12,session)=avg_steer_jerk(session); end end exp=0; non_exp=0; for i=1:37 if i==1|| i==7 || i==8 || i==9 || i==10 || i==17 || i==22 || i==23|| i==24 || i==25|| i==26 || i==27|| i==28 || i==29 || i==30||i==35||i==36||i==37 exp=exp+1; Steer_expert(exp,:,:)=STEER(i,:,:); else non_exp=non_exp+1; Steer_nonexpert(non_exp,:,:)=STEER(i,:,:); end end 87 for session=1:8 for i=1:12 [b(i,session),h(i,session)]=ranksum(Steer_expert(:,i,session), Steer_nonexpert(:,i,session)); end end speed=[70,80,85,90,95,100,105]; sp(1:7)=speed; sp(14:-1:8)=speed; for i=1:12 figure(i) set(gcf,'Color',[1,1,1]); title('EXPERIENCED vs NOVICES') for j=1:7 mx1(j)=mean(Steer_expert(:,i,j)); mx2(j)=mean(Steer_nonexpert(:,i,j)); pr25_1(j)=prctile(Steer_expert(:,i,j),25); pr25_2(j)=prctile(Steer_nonexpert(:,i,j),25); pr75_1(j)=prctile(Steer_expert(:,i,j),75); pr75_2(j)=prctile(Steer_nonexpert(:,i,j),75); end hold on pr1(1:7)=pr25_1; pr1(14:-1:8)=pr75_1; pr2(1:7)=pr25_2; end for i=1:7 figure(13) title('EXPERIENCED vs NOVICES') set(gcf,'Color',[1,1,1]); hold on scatter(Steer_expert(:,10,i),Steer_expert(:,2,i),'r') scatter(Steer_nonexpert(:,10,i),Steer_nonexpert(:,2,i),'g') end legend('EXPERIENCED','NOVICES') %% speed vs cones hit/rms driver_as_exp=[-10 -1 -1 -15 -18 -13 -2 -3 -1 -8 -17 -14 -8 0 -8 -8 -5 -2]; driver_as_nonexp=[1 -4 20 -4 -16 -4 29 0 -1 8 3 6 -3 6 2 12 6 -4 1]; driver_age_exp=[28 28 26 26 29 35 26 26 31 26 28 31 30 26 27 28 30 26]; driver_age_nonexp=[19 19 20 20 19 20 21 20 19 21 21 21 21 19 20 21 20 19 21]; pr2(14:-1:8)=pr75_2; [ph1,msg1]=jbfill(speed,pr25_1,pr75_1,[1 0 0],[1 0 0],0,0.1); [ph2,msg2]=jbfill(speed,pr25_2,pr75_2,[0 1 0],[0 1 0],0,0.1); plot(speed,mx1,'--r*','LineWidth',2) plot(speed,mx2,'--g*','LineWidth',2) title('EXPERIENCED vs NOVICES') legend('EXPERIENCED','NOVICES') plot(speed,pr25_1,'--ro'); plot(speed,pr25_2,'--go'); plot(speed,pr75_1,'--rd') plot(speed,pr75_2,'--gd') xlabel('SPEED (kmph)') end for session=1:8 eval(sprintf('SESSION%d(1:18,1:12)=Steer_expert(:,:,sessio n);',session)); eval(sprintf('SESSION%d(19,1:12)=mean(Steer_expert(:,:,s ession));',session)); eval(sprintf('SESSION%d(20,1:12)=std(Steer_expert(:,:,sess ion));',session)); eval(sprintf('SESSION%d(22:40,1:12)=Steer_nonexpert(:,:,s ession);',session)); eval(sprintf('SESSION%d(41,1:12)=mean(Steer_nonexpert(: ,:,session));',session)); eval(sprintf('SESSION%d(42,1:12)=std(Steer_nonexpert(:,:, session));',session)); eval(sprintf('SESSION%d(43,1:12)=b(:,session);',session)); 88 und=data.Y(1,10).Data; 3. Oval Track test l=length(X); %% OVAL TRACK DRIVING clear all close all count1=0; % LAP SELECTION %% DEFINING FILTERS [z,p,k]=butter(2,3/50); %% 3HZ LOW PASS FILTER [sos,g]=zp2sos(z,p,k); hd=dfilt.df2sos(sos,g); [z1,p1,k1]=butter(2,10/50); %% 10HZ LOW PASS FILTER [sos1,g1]=zp2sos(z1,p1,k1); hd1=dfilt.df2sos(sos1,g1); x1=40:1:75; y1=3; y2=-3; y3=153; y4=147; y_exp=0; y_nov=0; h=0; nh=0; angle=linspace(pi/2,3*pi/2,100); r1=78; r2=72; c1=75-r1*cos(angle); c2=75-r1*sin(angle); c3=75-r2*cos(angle); c4=75-r2*sin(angle); close all % hold on % plot(x1,y1) % plot(x1,y2) % plot(x1,y3) % plot(x1,y4) % plot(c1,c2,'--r') % plot(c3,c4,'--r') % yy=0; % n=0; %% LOADING DRIVER DATA for D=1:37 clearvars -except D hd hd1 count1 STEER y_nov y_exp h nh num_crash num_laps P=load(sprintf('%s%d%s','Driver_',D,'_Trial_highmu.mat')); data=P.(sprintf('%s%d%s%d','driver_',D,'_trial_0',41)); steer=filter(hd,data.Y(1,13).Data); steer_diff=diff(steer); steer_jerk=abs(diff(steer,3)); X=filter(hd1,data.Y(1,23).Data); Y=filter(hd1,data.Y(1,24).Data); velx=filter(hd1,data.Y(1,2).Data); vely=filter(hd1,data.Y(1,3).Data); velz=filter(hd1,data.Y(1,4).Data); accx=filter(hd1,data.Y(1,5).Data); accy=filter(hd1,data.Y(1,6).Data); accz=filter(hd1,data.Y(1,7).Data); t=1; n=2; lap(1)=1; while t0 && t50)&&flag==1 entry_curve1(i)=j; k=j; while (X(k)<100 && k50 && k0 && k25 exit_curve2(i)=k; else exit_curve2(i)=0; end end end end % CRASHES for i=1:length(exit_curve1) crash(i)=0; if exit_curve1(i)>0 for j=entry_curve1(i):exit_curve1(i) % if data.Y(1,19).Data(1,j)~=0.4 &&data.Y(1,20).Data(1,j)~=0.4 && data.Y(1,21).Data(1,j)~=0.4 && data.Y(1,22).Data(1,j)~=0.4 if data.Y(1,19).Data(1,j)<1 &&data.Y(1,20).Data(1,j)<1 && data.Y(1,21).Data(1,j)<1 && data.Y(1,22).Data(1,j)<1 crash(i)=1; end end end end for i=1:length(crash) % if crash(i)==0 && exit_curve1(i)>0 % if D==7||D==8||D==10||D==17 % figure(D) % end % if D==16||D==18||D==20||D==21 % % figure(2) % end % hold on %% subplot(2,1,1) % % plot(steer(entry_curve1(i):exit_curve1(i))) % % hold on %% subplot(2,1,2) %% %% plot(und(entry_curve1(i):exit_curve1(i))) % % end end num_crash(D)=histc(crash,1); num_laps(D)=length(crash); %% CALCULATING DRIVER DEPENDANT MEASURES for i=1:length(exit_curve1)-1 count=0; if entry_curve1(i)>0 && exit_curve1(i)>0 curvetime(i)=(-entry_curve1(i)+exit_curve1(i))/100; rms_acc(i)=sqrt(mean(accel(entry_curve1(i):exit_curve1(i)). ^2)); for j=entry_curve1(i):exit_curve1(i) if sign(steer_diff(j))~=sign(steer_diff(j+1))&&steer_diff(j)~=0 &&steer_diff(j+1)~=0&&abs(steer(j)-steer(k))>3 flag =0; p=j+1; while(flag==0&&p3 count=count+1; k=j+1; pos3(i,count)=j+1; end else p=p+1; end end end end if count>0 SRR(i)=count/((pos3(i,count)-pos3(i,1))*.01); else SRR(i)=count; end con=0; for m=(entry_curve1(i):exit_curve1(i)) if und(m)==1 con=con+1; end end unders(i)=con/(-entry_curve1(i)+exit_curve1(i))*100; overs(i)=100-unders(i); sum1=sum(abs(steer_diff(pos3(i,1):exit_curve1(i)))); sum2=sum(abs(steer_jerk(pos3(i,1):exit_curve1(i)))); max1=max(steer_jerk(pos3(i,1):exit_curve1(i))); avg_steer_speed(i)=sum1/(0.01*(pos3(i,1)+exit_curve1(i))); avg_steer_jerk(i)=sum2/(0.000001*(pos3(i,1)+exit_curve1(i))); max_steer_jerk(i)=max1/0.000001; 90 d=0; for j=entry_curve1(i):exit_curve1(i)-1 k=j-entry_curve1(i)+1; d=d+sqrt((X(j+1)-X(j)).^2+(Y(j+1)-Y(j)).^2); X_diff=75-X(j); Y_diff=75-Y(j); R1(i,k)=sqrt(X_diff.^2+Y_diff.^2); Theta(i,k)=atan(Y_diff/X_diff); V=sqrt(velx(j).^2+vely(j).^2+velz(j).^2); ACC=sqrt(accx(j).^2+accy(j).^2+accz(j).^2); TR3(i,k)=V.^2/ACC; CURV3(i,k)=ACC/V.^2; end dis(i)=d; SORT1=sort(CURV3(i,:),2,'descend'); SORT1=SORT1(:,1:5); SUM(i)=sum(SORT1,2); end end yy=0; n=0; for k=1:length(crash) if crash(k)==0 % hold on if entry_curve1(k)>0 && exit_curve1(k)>0 n=n+1; angle=linspace(pi/2,3*pi/2,100); xi=-2:0.1:152; xa=Y(entry_curve1(k):exit_curve1(k)); ya=X(entry_curve1(k):exit_curve1(k)); yi=interp1(xa,ya,xi); for len=1:size(yi,2) if isnan(yi(len))~=1 dev(k,len)=sqrt((75-yi(len))^2+(75-xi(len))^2)72; end end yy=yy+yi; flag=1; for r=1:length(yi) if yi(r)>100&&flag==1 st=r; flag=2; end if yi(r)<100&&flag==2 en=r; flag=1; end end figure(D) subplot(2,1,1) hold on plot(yi(st:en),dev(k,st:en)/dev(k,st)) subplot(2,1,2) hold on plot(X(entry_curve1(k):exit_curve1(k)),Y(entry_curve1(k):e xit_curve1(k))) %plot(yi,xi) %plot(X(entry_curve1(k):exit_curve1(k)),Y(entry_curve1(k ):exit_curve1(k))) end end end path_dev=mean(dev(:,200:1400),2); path_std=std(dev(:,200:1400)'); yy=yy/n; % plot(yy,xi,'k') for i=1:length(exit_curve1)-1 if crash(i)==0 && entry_curve1(i)>0 && exit_curve1(i)>0 && curvetime(i)<30 count1=count1+1; STEER(count1,1)=D; STEER(count1,2)=curvetime(i); STEER(count1,3)= avg_steer_speed(i); STEER(count1,4)= avg_steer_jerk(i); STEER(count1,5)=rms_acc(i); STEER(count1,6)=SRR(i); STEER(count1,7)=unders(i); STEER(count1,8)=overs(i); STEER(count1,9)=path_dev(i); STEER(count1,10)=path_std(i); end end end %% SPERATION INTO EXPERIENCED AND NOVICES exp=0; nexp=0; for i=1:size(STEER,1) if STEER(i,1)==7 ||STEER(i,1)==8 ||STEER(i,1)==9 ||STEER(i,1)==10 ||STEER(i,1)==17 ||STEER(i,1)==22 ||STEER(i,1)==23|| STEER(i,1)==4||STEER(i,1)==15||STEER(i,1)==26||STEER (i,1)==1|| STEER(i,1)==27||STEER(i,1)==28||STEER(i,1)==29||STEE R(i,1)==30|| STEER(i,1)==35||STEER(i,1)==36||STEER(i,1)==37 if STEER(i,7)>3 exp=exp+1; steer_exp(exp,:)=STEER(i,:); end else nexp=nexp+1; steer_nexp(nexp,:)=STEER(i,:); end end 91 %% CALCULATING THE AVERAGE FOR THE TWO GROUP OF DRIVERS start=1;D=2;m=0; for i=1:size(steer_nexp,1) if(steer_nexp(i,1)~=D || i==size(steer_nexp,1)) figure(2) hold on m=m+1; mean_nexp(m,1)=D; mean_nexp(m,2:10)=mean(steer_nexp(start:i,2:10)); mean_nexp(m,11)=std(steer_nexp(start:i,9)); plot(D,mean_nexp(m,2),'g') hold on D=steer_nexp(i,1); start=i; end end start=1;D=1;m=0; for i=1:size(steer_exp,1) if(steer_exp(i,1)~=D || i==size(steer_exp,1)) figure(2) m=m+1; mean_exp(m,1)=D; mean_exp(m,2:10)=mean(steer_exp(start:i,2:10)); mean_exp(m,11)=std(steer_exp(start:i,9)); plot(D,mean_exp(m,2),'r') hold on D=steer_exp(i,1); start=i; end end %% RANKSUM T-TEST for i=1:11 p(i)=ranksum(mean_exp(:,i),mean_nexp(:,i)); end 92