Transcript
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
Transportation safety research applications utilizing high-fidelity driving simulation G.S. Watson, Y.E. Papelis & L.D. Chen National Advanced Driving Simulator, The University of Iowa, USA Abstract According to U.S. government statistics, approximately 42,000 lives are lost annually in automobile accidents on U.S. highways. A large percentage of these accidents can be partially attributed to human error. Any technology that improves understanding of human driving performance can have a drastic effect on reducing fatahties and other ancillary societal costs. One technology with such potential is high-fidelity driving simulation. The US. Department of Transportation's National mghway Traffic Safety Administration constructed the National Advanced Driving Simulator, the highest fidelity ground vehicle simulator in the world, to investigate humamcenterecl issues as they relate to driving safety. The simulator's primary mission is to investigate causes of collisions, with the goal of reducing fatalities on U.S. roadways. This paper overviews its design, core capabilities, and key applications and dtscusses how they may influence safety. The National Advanced Driving Simulator is operated by The University of Iowa on a self-sustaining basis.
Introduction Through a competition administered by the National Science Foundation, the U.S. Department of Transportation awarded The University of Iowa (UI) the host site of the National Advanced Driving Simulator (NADS) in 1992. The NADS mission is to conduct research that will lead to a better understanding of the complex driver-vehicle-roadway interaction in critical driving situations and to a reduction in the number of traffic-related deaths and injuries on the nation's highways. The NADS design goals [l] were state-of-the-art simulation fidelity, simulation of vehicle response to limit conditions, user-friendly operation and interface, flexibility and rapid configuration, maximum safety to participants and
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
staff, highly efficient two-shift operation, low-cost experiment rehearsal and debugging, and minimized simulator sickness. NADS is a national shared-use research facility owned by the National Highway Traffic Safety Administration (NHTSA) and operated by the UI. It delivers a high-fidelity simulator with a large motion envelope and a fully immersed visual environment, a software library, a simulation development module (SDM), and four full-sized, hlly instrumented vehicle cabs. The NADS is housed in a 40,000-ft2 facility located on the U1 Oakdale Campus. The following sections outline the detailed motivation for the NADS along with technical information on the NADS system, including scenario authoring tools and the scenario control module, data reduction and verification tools, the NADSdyna vehicle dynamics software, and potential research areas particularly suited for this unique research device.
NADS uses Simulators are useh1 because they provide a safe, reproducible, and often economical environment to train, assess, and study individuals in situations that are otherwise impossible or impractical. Simulators facilitate easy modification of test parameters, exact control of experimental conditions, minute levels of performance measurement, and the ethical induction of hazardous conditions. The high-fidelity features of the NADS are particularly well suited to applications where a highly realistic and irnrnersive environment is required and where little or no practice time is required before driving performance in the simulator relates closely to that in the real world. Such applications include the study of reduced visibility and driver impairment due to fatigue, drugs, alcohol, or medical conditions such as dementia. The practice time required for the NADS is similar to that required to become familiar with a rental car. The close performance match between the NADS and the real world can be attributed to the realism of the virtual environment. The perceived similarity extends to many driving maneuvers, including turns, braking, and obstacle avoidance. In lower fidelity environments, the driver is often restricted to driving straight because of the difficulty of learning or performing more demanding maneuvers. The NADS is also valuable in situations that require manipulation and complete correspondence of real-time multi-body vehicle dynamics, vehicle control response, motion cueing, and visual scenes. The ability to accurately measure driver performance and vehicle handling is closely associated with the ability to precisely model the vehicle while providing corresponding perceptual cues to the driver. These features make the NADS robust for studying advanced technologies aimed at better vehicle control and automation. Accurate roadway modeling coupled with the ability to provide complex interactions with other vehicles makes NADS particularly well suited for complex driving situations involving driver distraction, impairment, and aging. Furthermore, the infrastructure supports rapid integration of additional hardware for studying advanced in-vehicle technologies such as collision warning systems, adaptive cruise control, and in-vehicle telematics systems.
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
Overall NADS system description NADS modules provide a high level of fidelity for driving simulation that helps researchers understand complex driver, vehicle, and roadway interaction during critical maneuvers. During simulation, a NADS cab is placed in a 7.3-m (24-ft) dome (see Figure 1) that is attached to a motorized turntable on a hexapod (sixlegged) motion platform. This platform is mounted on a large track capable of reproducing extensive longitudinal and lateral motion cues to the driver. This configuration provides the substantial accelerations required for realistic passing and braking maneuvers, while high-frequency actuators attached to the vehicle cab accurately reproduce subtler sensations such as gravel, tar strips, and potholes. Multiple projectors and a surround sound system immerse the driver in a realistic visual and audio environment generated in real-time from programmable virtual worlds that provide detailed terrain and roadways, intelligent traffic, a variety of traffic signals and signs, pedestrians, weather conditions, and other interactive features. In addition to the software that runs the NADS system, numerous other software packages facilitate the development of virtual environments, authoring of scenarios, and post-processing of data. The goal of these software tools is the rapid development of simulator configurations with specifications that match research requirements. The ability to rapidly develop new scenes and scenarios is critical for the effective use of the NADS. The remainder of this section outlines the NADS subsystem and software tools. NADS subsystems The visual system is driven by an image generator that produces real-time textured imagery, which in turn is projected on the 7.3-m (24-ft) dome through fifteen projectors. Each projector has a maximum resolution of 1024 X 768 pixels, although both higher and lower resolution modes are used depending on where the scene is projected. A high-resolution ( l .l arc minutes per optical line) inset is projected immediately ahead of the driver to ensure maximum fidelity imagery projection. The visual system has a sustained polygon throughput of 21,000 polygons per frame at a 60 Hz update rate. The driver is provided with realistic fields of view, including rearview images viewed through the vehicle mirrors. The motion system provides the largest translational motion envelope ever developed for a driving simulator-64x64 feet. It also provides a continuous 330degree turning yaw angle in both clockwise and counter-clockwise directions, pitch and roll, and high-frequency cues that duplicate motion over the full range of driving maneuvers. The motion system provides an instantaneous acceleration of +l- 0.6 g in translational directions and +l- 1 g in the vertical direction. The maximum speed of the dome is 6.1 m (20 ftls). The maximum frequency and displacement of the high-frequency actuators is 20 Hz and +/- 0.5 cm (0.2 in).
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
196 Urban Transport and the Environment in the 21st Century
Figure 1: Photograph of the NADS dome. The control feel systems provide the driver with realistic feedback on steering, brakes, clutch, transmission, and throttle. The control feel system is capable of representing automatic and manual control characteristics such as power steering, existing and experimental drive trains, anti-lock braking systems, and cruise control. Obtaining variable pedal feel is possible though the use of active feedback mechanisms, as opposed to passive systems, for all pedals. The audio system provides three-dimensional sound sources coordinated with other sensory systems. The auditory system has a bandwidth from 15Hz to 20 kHz and a signal-to-noise ratio of 9533. The audio database includes sounds emanating from highway surfaces, high-density multiple-lane traffic, the vehicle during operation (engine, brake, and wind noise), and the roadway due to changes in the synthetic weather conditions. In addition, entities of the virtual environment have their own passive and active audio signatures. Passive audio signatures are played using high fidelity 3D audio reconstruction when an entity is near enough to the driver to be heard. Active audio signatures are played when a virtual environment entity comes in contact with the driver's vehicle. The NADS cabs consist of actual vehicle cabs configured to fit inside the dome. Video cameras in the cab can be used to capture simultaneous images of the driver, foot and hand positions, and other configurable views. The cabs have a full range of vehicle instrumentation interfaces. Four cabs are provided: a Chevy Malibu, a Ford Taurus, a Daimler Chrysler Cherokee, and a Freightliner Century truck. Cabs are designed and constructed to allow rapid interchangeability, with a typical exchange time of eight hours. The SDM is a NADS-like simulator for off-line development of experiments and virtual environments. The SDM contains all the basic elements of the NADS, except motion, and uses only three projectors. Duplicate hardware resources make it possible to operate the NADS and SDM simultaneously, enabling experimenters to cost-effectively preview scenarios and drive through scenes to assure themselves that the experiment is well structured before carrying out
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
experiments on the full NADS system. The SDM can also be used as a fixed-base simulator for data collection. A NADS software library contains various programmable components necessary for the operation of the system, along with tools that support the creation of new components. The library contains vehicle dynamic models, audio signatures for various effects, visual databases, simple standard scenarios, and default post-processing capabilities. The library is continuously enhanced with the addition of new capabilities that represent work done for various studies. Due to their critical role in the NADS, some components of the library along with the tools are further described below. Vehicle dynamics The dynamics used on the NADS are built upon a UI-developed general-purpose dynamics code, NADSdyna. Each vehicle model used in the NADS consists of three components: (1) major mechanical vehicle components such as chassis, suspension, and steering linkages, (2) vehicle subsystems such as vehicle steering, braking, engine, powertrain, tires andlor tracks, displays, and controls, and (3) environmental subsystems such as tire-soil interaction, aerodynamics, and other environmental effects. NADSdyna uses the multi-body approach for modeling the main vehicle moving mechanical parts. A library of vehicle subsystem models developed and validated by the Vehicle Research Test Center [2-61 and corresponding to the standard cabs is available. For new models, due to the varying nature of vehicle subsystems and depending on the nature of the simulator experiments to be conducted, parametric variations of the available models, or in some cases complete remodeling tasks, can be conducted. In addition to the four basic models, hybrid-electric component models for batteries, electric motors, motor controllers, and generators have been formulated and demonstrated. Scenario authoring tools and scenario control Many factors contribute to the irnrnersive experience of a simulator, but when the subject under study requires interaction with other vehicles, it is important that the virtual environment includes a microscopic traffic simulation model that generates traffic to populate the simulator's virtual environment. The NADS uses a sophisticated scenario definition and control system that consists of software tools that can be used to author, rehearse, and test scenarios before executing them on the simulator, along with a runtime engine that executes scenarios in real time. The scenario definition and control system consists of a scene authoring component, a scenario authoring and rehearsal tool, and the runtime system. The scene-authoring component produces the static elements of the virtual envtronrnent. The most common and efficient method for producing visual databases involves the Tile Mosaic Tool, which allows users to combine pre-built tiles into a larger environment. The library contains approximately 70 tiles, including highways, mountainous roads, and residential, rural, and industrial areas. New tiles can be developed for imaginary or geo-specific locations. Each
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
tile contains a correlated version of the road network to be used by the scenarioauthoring tool and runtime system. The correlated database contains a detailed description of the road network. A real-time Applications Programming Interface allows programmatic access to the Correlated Virtual Environment Database [7] that is used by the scenario authoring system and runtime engine. The scenario-authoring component, facilitated by the Interactive Scenario Authoring Tool (see Figure 2), allows users to construct a scenario to be experienced by the simulator driver. A scenario includes the simulator initial conditions, traffic light configurations and timing, and a specification of the traffic to be encountered. Traffic can be either explicitly specified, where each vehicle is placed at an initial position and allowed to navigate freely around the database, or interactively managed by the traffic manager. Traffic can be placed either as autonomous objects, each of which uses a comprehensive driver model, or as deterministic route followers that are computationally cheap but exhibit no behavior other than following a geometric path and slowing down to avoid collisions. Rules can be defined though the use of triggers, sources, and other coordinators that allow the orchestration of deterministic events despite the general randomness of the traffic. An environment model allows the arbitrary specification of weather, visibility and similar conditions, and fine grain control of time of day. Papelis, Ahrnad, and Schikore [8] provide additional details on scenario authoring capabilities. The runtime scenario system consists of the traffic simulation module and the software that integrates it into the simulator. The scenario control system uses a microscopic traffic simulation model in which the individual behavior of each vehicle is modeled and simulated. The particular model used on the NADS [9] incorporates various behaviors whose purpose is to produce realistic interaction with the simulator driver [10,11], in addition to the typical lane tracking and following behaviors. The traffic simulation system is built using the Hierarchical Concurrent State Machine model [12], which allows easy addition of new behaviors. A key capability of the traffic simulation module is the ability to satisfy the contradicting requirements of randomness and control. Data reduction and verification It is not atypical in a driving simulation research study to spend half or more of the available project time verifying, reducing, and analyzing the data collected during the simulator runs. In order to streamline this process, the NADS uses a set of in-house tools that support data reduction, verification, and visualization of simulator data. These tools allow easy specification of data reduction procedures to make it simpler to identify technical blunders in the collection or interpretation of data, to reduce delays in delivering results, and to allow experimenters the ability to easily modify data reduction parameters and try what-if scenarios. They consist of three major components: the front-end graphical user interface (see Figure 3), the execution engine, and the data reduction server. Papelis et al. provide a comprehensive description of the data reduction tools [13].
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
Figure 2: Main Interactive Scenario Authoring Tool window.
Figure 3: Data reduction specification tool. The front-end graphical user interface allows users to create a data reduction specification using a visual dataflow language defined specifically for this purpose. Users create a dataflow graph that, when executed, performs transformations on the raw data collected at the simulator. The execution engine executes the data reduction graph and performs the necessary actions. This software has unique features that target application problems typical in driving simulation, one of which is the massive amount of non-changing data. The data reduction server is a custom software package designed to work with the execution engine and the graphical user interface tool. It stores all parameters and descriptions associated with a simulator data collection and provides authorized access to it. The server is aware of the nature of the data; for example, it keeps associations of binary files with specific simulator subjects or runs. In addition, the server can store video or arbitrary documents, which can also be associated with a particular experiment, subject, or simulator run. Documents can be organized in arbitrary categories within a given experiment, and flexible data
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
access rights can be associated with the server itself, data associated with an experiment, or data associated with a given category. The data reduction server simplifies logistics because it eliminates the need to maintain data on portable storage or on individual user workstations. It also reduces security issues associated with numerous copies of data on CDs or other removable media.
Recent upgrades In response to project requirements, the NADS has recently been upgraded to include a real-time non-intrusive eye tracking system and sophisticated digital video recording equipment. The eye-tracker is entirely camera-based, does not require headgear, and provides real-time 3D head position, eye-gaze direction, and blink detection at a 60 Hz rate. It tracks primary variables under head rotations of up to 90 degrees in each direction and is very robust even under the simulator's low-light conditions. Collected data is synchronized with other simulator variables, offering unique insight into driver activities, attention, and vigilance during simulation. The video recording equipment consists of four real-time MPEG2 recorders hooked to four digital VCRs, providing the option of recording to tape andior hard disks. Approximately 24 discrete video signals are available, including all output channels of the image generator, in-cab video cameras (face, feet, overthe-shoulder, etc.), in-dome video cameras, and cameras overseeing the motion bay. Up to four images can be multiplexed into a single video stream, and a selectable display of real-time simulation variables can be overlaid on any stream. Recording is synchronized with simulator operation, so it is possible to correlate a particular point on any video with the corresponding time within the simulation.
Current applications A variety of projects are underway or planned for the NADS. Although details vary, projects can be classified among these areas: driver impairment; driver distraction, advanced technologies, and automation; and core simulator technologies.
Driver impairment Driver impairment is a critical safety issue that can be effectively studied at the NADS. Highway safety researchers can develop methods for assessing impairment and fitness to drive for special populations, drivers under the influence of alcohol and prescription or non-prescription drugs, and drivers with medical or physical conditions that affect cognitive processes or physical response. Assessment techniques used to date have largely involved lower fidelity simulations or laboratory tests believed to relate to driving behavior. Many of these studies have had limited generalizability because of limited representation of experimental tasks to a real driving environment with complex driving tasks. The NADS provides a safe and interactive environment for measuring driving performance and eye movements in conditions and situations
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
that are associated with typical crashes involving impaired drivers. Specific studies will focus on the relationships between driver task demands, driver situational and environmental demands, and diurnal variation and fatigue when crossed with driver demographics, drinking practice, age, gender, and various levels of blood alcohol concentration (BAC). Drug studies will compare the effects of various drugs on driving performance. Proposed studies to investigate fitness-to-drive will examine such issues as the effect of sleep apnea, cervical collar use, and ophthalmologic devices on driving performance.
Driver distraction, advanced vehicle technologies and automation As our roads become more crowded, as drivers spend more time on the road, and as driver multitasking becomes more prevalent, driver distraction issues are a growing safety concern. Driver distraction generally involves diverting attention and resources away from the primary driving tasks to other secondary or nondriving tasks. Typical forms of driver distraction involve the use of wireless phones, interactions with other passengers, or the use of competing in-vehicle technologies. In the NADS, researchers can develop complex driving environments and intricate interactions with other vehicles to determine how driver demands affect driver performance. Further, researchers can integrate advanced technologies such as collision warning systems, adaptive cruise control, and integrated telematics systems that are intended to assist drivers in order to optimize design and interfaces of individual or multiple technologies. Specific studies will focus on how various wireless phone interfaces and different types of conversation content influence driving performance in a variety of driving scenarios. Other studies will investigate how drivers interact with various invehicle technologies in situations commonly associated with crashes to determine how these systems improve crash mitigation for a wide variety of drivers. Proposed research will also investigate the effect of automation and futuristic automotive technologies.
Core simulator technology In addition to human-centered testing, the NADS can provide a unique platform against which to measure the effectiveness of various simulation technologies. Comparisons with fixed based simulators, or with simulators with smaller motion systems, can provide unique data on the usability boundary for lower fidelity devices. Projects currently underway will leverage NADS technology and incorporate it into fixed-base simulators. Future plans include the direct comparison of driving performance among simulators that are based on the same software architecture but have different fidelity cueing subsystems.
References [l] Brewer, H.K., National Advanced Driving Simulator, Paper presented at the Global Automotive Safety Conference, Society of Plastics Engineers, February 5 , 2001.
Transactions on the Built Environment vol 64, © 2003 WIT Press, www.witpress.com, ISSN 1743-3509
[2] Garrott, W.R. et al., Methodology for Validating the National Advanced Driving Simulator S Vehicle Dynamics (NADSdyna), SAE Paper 970563, February 1997. [3] Chrstos, J.P. & Grygier, P.A., Experimental Testing of a 1994 Ford Taurus for NADSdyna Validation, SAE Paper 970563, February 1997. [4] Salaani, M.K., Heydinger, G.J. & Guenther, D.A., Validation Results from Using NADSdyna Vehicle Dynamics Simulation, SAE Paper 970565, February 1997. [5] Salaani, M.K. & Heydinger, G.J., Powertrain and Brake Modeling of the 1994 Ford Taurus for the National Advanced Driving Simulator, SAE Paper 981 190, February 1998. [6] Salaani, M.K., Guenther, D.A. & Heydinger, G.J., Vehicle Dynamics Modeling for the National Advanced Driving Simulator of a 1997 Jeep Cherokee, SAE Paper 1999-01-0121, February 1999. [7] Papelis, Y. & Bahauddin, S., Logical Modeling of Roadway Environments to Support Real-Time Simulation of Autonomous Traffic, First Workshop on Simulation and Interaction in Virtual Environments, Iowa City, IA, USA, pp. 62-71, 1995. [8] Papelis, Y,, Ahrnad, 0. & Schikore, M,, Scenario Definition and Control for the National Advanced Driving Simulator, Paper presented at the Conference on Enhanced Safety of Vehicles. Amsterdam, The Netherlands, 200 1. [9] Papelis, Y. & Ahmad, O., A Comprehensive Microscopic Autonomous Driver Model for Use in High-Fidelity Driving Simulation Environments, Paper presented at the Transportation Research Board Conference, Washington, D.C., USA, 2001. [IOIAhmad, O., Papelis, Y., Bulusu, S. & Gade, V., Automatic Learning by Autonomous Driver Agents as Applied to Performing Realistic Lane Change Maneuvers, ISHFproceedings, Sapporo, Japan, pp 109- 1 14,200 1. [l lJAhmad, 0. & Papelis, Y., An Autonomous Driver Model for the Overtaking Maneuver for Use in Microscopic Trafic Simulation, Paper presented at the Driving Simulation Conference, Paris, France, 2000. [12]Cremer, J., Kearney, J. & Papelis, Y., HCSM: A Framework for Behavior and Scenario Control in Virtual Environments, ACM Transactions of Modeling and Computer Simulation, 1995, Vol. 5, No. 3, pp. 252-267. [l31Papelis, Y., Schikore, M., Watson, G. & Gates, T., A Prototype Toolset for Rapidly Reducing and Reviewing Driving Simulator Data, Paper presented at the l'' Human-Centered Transportation Simulation Conference, Iowa City, IA, USA, 200 1.
