Transcript
Feasibility Study for the Automation of Commercial Vehicles on the Example of a Mobile Excavator Carsten Hillenbrand1 , Daniel Schmidt1 , Nureddin Bennett2 , Peter Bach3 , Karsten Berns1 , and Christian Schindler2 1
Robotics Research Lab at the Department of Computer Sciences, University of Kaiserslautern, P.O. Box 3049, 67653 Kaiserslautern, Germany 2 Chair of Design in Mechanical Engineering, University of Kaiserslautern, Gottlieb-Daimler-Strasse 42, 67663 Kaiserslautern, Germany 3 VOLVO Construction Equipment GmbH & Co. KG, 54329 Konz, Germany
Abstract. The automation of mobile working equipment used in farming, forestry and the construction industry has great market potential. Tasks of a highly complex nature could be executed without intervention of the operator. Despite the potential significance of this topic, only very few results of international research projects in robotics have heretofore been applied in the serial production of commercial vehicles. The joint project AMoBa (Autonomer Mobiler Bagger, Autonomous Mobile Excavator) of Kaiserslautern University and Volvo CE aims to transfer knowledge and experience from the field of robotics to the area of mobile excavators. The objective is the realization of fully autonomous operation in basic construction activities, such as terrain modeling or trenching. For this purpose a prototype is being prepared by modifying an 18t Volvo Mobile Excavator to meet the project’s requirements. It is being equipped with electronic interfaces to control all relevant functions; i.e. the hydraulic arm cylinders, thereby employing sensor devices for the acquisition of the kinematics (i.e. travel sensors or angular sensors) and sensors for the acquisition of the environment surrounding the robotic excavator. For example, 2-D laser scanning devices and 3D cameras. Mechanical and hydraulic simulation models are being created to enable studies of suitable algorithms for trajectory control of the excavators’ kinematics. Algorithms concerning the global task of processing the automation task are developed and tested in the software framework MCA2 before they are ported to the prototype. The MCA2 framework combines the simulation of the excavator and its environment, including the simulation of environment sensors. On completion of this project, the algorithms developed in the simulation will be verified on the prototype.
1
Background
The importance of autonomous machines and assistance functions is permanently growing. Leading car manufactures and research institutes attend challenges for fully autonomous cars. In the field of mobile machines autonomous guidance systems for agricultural vehicles are already available on the market.
Referring to this, the visionary objectives are mobile construction machines that work like a robot with versatile benefits. For instance the operating costs could be lower due to less manpower requirements and less non-working time. The machines could be used in rough working environments without endangering any operator. The operating conditions could be adjusted in a way that fuel efficiency or material damage is reduced to a minimum. Regarding the humanmachine interface a landscape architect or a construction engineer could design a given environment on a PC. Then, the construction machines could start their work autonomously exactly reproducing the architects ideas. Apart from these visions autonomous machines will only have market opportunities when they demonstrate their profitability in customer use and when they have a safety standard that is comparable to ordinary machines or even higher. With this background, the pilot project investigates how autonomous functions could be realized on a mobile wheeled excavator. This type of machine has been chosen for the project because of the wide range of applications that are even beyond typical earth moving machinery tasks. Excavators are comparable with industrial automation robots. The kinematics of the digging apparatus is quite similar and it can be equipped with a variety of tools. Hence, excavators can be seen as mobile machining tools.
2
Project AMOBA
The joint project AMoBa (Autonomer Mobiler Bagger, Autonomous Mobile Excavator) of Kaiserslautern University and Volvo CE aims to realize fully autonomous operation of a mobile excavator. Kaiserslautern University is involved with two chairs, the Robotics Research Lab RRLab and the Chair of Design in Mechanical Engineering KIMA. The project AMoBa is finacially supported by the ”Stiftung Rheinland-Pfalz f¨ ur Innovation”. Started in August 2008, it will be supported until end of 2010. The specific task is to accomplish autonomous terrain modeling and trenching with a given VOLVO EW180, an 18 ton state of the art mobile excavator. RRLab has extensive knowledge in mobile robotics using alternative approaches like behavior-based control. One very successful example is the autonomously driving vehicle RAVON. In this project, existing knowledge in robotics of RRLab is transfered to an industrial application. KIMA supports the project with its experience in commercial vehicles, i.e. the implementation of electro-hydraulic interfaces and accurate simulation models of the mechanical and hydraulic systems involved. The project formulation is a first approach into the wide-ranging challenge of autonomous machinery for commercial purposes. The manual controls are to be altered for digital computer control. However, the current state of the art in mobile excavators is based on pure hydraulic actuation, with very little electronics involved. The mechanical stresses on excavators in daily use are massive. At the same time, the requirements for reliability are extremely high. Heretofore, suitable sensor systems are to be found to measure the current state of internal
parameters. Even more important is the acquisition of environmental information with regards to the area surrounding the excavator. All sensory information has to be read out and evaluated in real-time. As the next step, suitable control algorithms for the two given digging tasks have to be developed. This involves the implementation of the pure digging process, bucket-ground interaction, and finally the execution of the main objective of performing defined landscaping and trenching. In order to concentrate on the main objective, the driving operation of the excavator has been excluded. Within this project, the excavator will be standing on its’ outriggers. The environment surrounding the machine will be free from moving objects including human beings.
3
The challenge for robotics
Today’s standard industry robots operate with high precision at high velocities in the production process. Unfortunately, this is only possible as their complete motion is pre-calculated, including velocities and accelerations, and stored as a fixed program. As no variance is allowed in the plan, a static environment becomes necessary. The robots are surrounded by enclosed cages in order to prevent human beings from disturbing the manufacturing process. However, the situation in outdoor robotics is totally different concerning the environment. All information about the perimeter and the global position is unknown and has to be aquired. Disturbances are very likely to occur. In this case an appropriate solution for controlling the excavation process is required. Related work concerning the control of autonomous excavators [1] uses task centered planning structures defining goals for specific trench excavation tasks based on real operator behaviors [2]. Goals are divided into different activities. The planning structure is capeable of creating working trajectories ensuring smooth bucket movements. An additional strategy is implemented for the removal of obstacles inside the excavation space. As no complex sensor evaluation take place, the strategies are only functional up to a limited obstacle size. Others [3,4,5] can be described as parametrized scripted joint control, based on existing knowledge about excavation strategies. Here, a central planner creates the control values from excavation to dumping using a set of parametrized sequential scripts. However, in case of unexpected disturbances, script parameters have to be recalculated, which is likely to lack performance concerning external disturbances. Furthermore, scripts cannot be paused at arbitrary times, which leads to undesired repetitions of scripted steps or a complete maneuver.
4
Adaptive concept for outdoor robots
Due to disturbances in the area of outdoor robotics, an appropriate solution for controlling the highly dynamic excavation process is required. As behavior-based approaches have shown their suitability for various applications [6], the same can be expected for the field of construction equipment. Here, the behavior based
iB2C architecture [7] is used which is implemented in the Kaiserslautern branch of the Modular Controller Architecture MCA21 .
4.1
Behavior Based Control Architecture
Behaviors can be described as controlling units, which try to reach and to maintain a desired goal continuously, like a specific boom elongation or a turning velocity. As the individual goals may overlap, the behaviors, which are usually connected in a behavior network, may stimulate or inhibit others. The coordination of directly competing behaviors is accomplished by so-called fusion behaviors having the same interface as basic behaviors. A further hierarchical level can be introduced by arranging several behaviors in a behavioral group which works as a behavior itself again. The development process for iB2C starts with the top-down design stage where the given task is decomposed into sub tasks. It is described by means of the excavation of a trench, which can be partitioned into the following elements: 1. Make a surface scan with the laser scanners and identify a target excavation position. 2. Approach the target position with the bucket. 3. Scratch the surface deep enough to dig out the desired amount of soil. 4. Move the bucket to a target dumping position. 5. Dump the soil onto a pile at a given position. Each of these elements can be decomposed according to the degree of freedom they influence. These are: – – – –
Turn the torso angle (e. g. surface scanning) Adjust the bucket extension (e. g. approaching the excavation point) Adjust the bucket height (e. g. digging deep) Change the bucket pitch angle (e. g. excavation or dumping of soil)
Removing soil from the surface is performed similar to the strategy of a human driver, depicted in figure 1. First, the bucket vertically penetrates the surface until it reaches a depth of around 20 cm. Then, a scraping behavior is achieved by a combination of adjusting the bucket extension and the bucket angle while the bucket height and the torso angle are kept. As a whole, the process is performed by four behaviors influencing the required degrees of freedom (dof) of the bucket. 1
MCA2 is a framework for robot prototyping and development. Here the KL branch builds the background for the system. Resources and information how to use and install the system can be found under http://rrlib.cs.uni-kl.de/mca2-kl.
Fig. 1. Idealized desired trajectory of the bucket during the excavation process. First the bucket is lowered until the desired value of 20 cm is reached. Afterwards, it is permanently pulled and turned become to constantly decrease the boom length and adjust the bucket angle.
4.2
Environment perception
In order to obtain information regarding the environment surrounding the excavator, Laser scanning devices are used. Their 3D point clouds are permanently updated during the excavation process and deliver the basis for the surface evaluation algorithm, visualized in figure 2. It uses one two-dimensional grid for each point cloud and evaluates the average value of scanned points per cell. If no point in the cell can be found, the surrounding cells are used to build an average value where empty cells without surrounding filled ones will get a height in z direction of zero. Afterwards, the two evaluated grids for the actual and the desired surface are compared. A third height difference grid is built by subtracting the desired surface grid from the actual surface grid. To find possible positions for excavations areas in the size of the bucket, around 1.5 m2 in the actual case, with a minimum excavation depth of 20 cm, are located. The one with the most surrounding space and the deepest excavation depth will be chosen as excavation area.
5 5.1
Toolchain Mechanical Simulation
For the development of appropriate control algorithms, extensive simulation models are required. This involves the simulation of the mechanics and the actuators and drives of the excavator. The mechanics are modelled in a multi-body-
z Actual surface
y
Excavation areas x Chosen area
Desired surface
Fig. 2. Graphical representation of the surface distance evaluation algorithm using two grid structures for the actual surface and the desired surface. The four possible excavation areas are identified and the first area is chosen as it is surrounded by eight green fields and has the highest excavation depth, i.e. the red area at this position is high.
simulation environment. In this project the models are created in MSC.Adams2 . The drives and actuators used in today’s excavators are realized with highly sophisticated hydraulic systems. In the case of the EW180B the hydraulics incorporate load sensing technology which provides load-independent movements of the digging arm. The hydraulic system is modeled in LMS.Amesim3 , a 1-D multi domain system simulation software. Both mechanical simulation and hydraulic simulation are then connected through co-simulation. This final model makes it possible to develop optimized control algorithms for trajectory control of the digging arm, creating the required interface between computer software and the excavator. 5.2
Environment Simulation
Motivated by security reasons and therefore to prevent harm from human beings, buildings or the excavator itself, a safe test environment is required. The MCA2 framework is used to build up an interactive, real-time simulation of the excavator’s shape and it’s environment. Furthermore it allows the representation and testing of control and perception algorithms including sensor devices. As the excavator is a highly dynamic system, a physics engine4 is used to simulate the 2
3
4
MSC.Adams is a commercial software for the dynamic simulation of multi-bodysystems. For further information refer to www.mscsoftware.com. LMS.Amesim is a software package for 1 dimensional transient simulation of multiple domain systems. For further information refer to www.lmsgermany.com The Newton Game Dynamics physics engine is an integrated solution for real time simulation of physics environments. Additional information can be found under http://newtondynamics.com.
excavator masses and joints and produces a quite realistic behavior of the whole machine. Interaction with other dynamic elements like walls or other machines is also included. As sensor systems like laser scanners or stereo vision cameras can be simulated too, appropriate mounting positions and algorithms for environment perception can be safely evaluated. A screen-shot of the simulation including a three dimensional surrounding area is shown in figure 3.
Fig. 3. Screen-shot from the test simulation showing the excavator in the 3Denvironment. The yellow vertical bars represent artificial landmarks which are used for determining the actual excavator position via triangulation from the perceived sensor data.
5.3
Hardware Implementation in EW180B Excavator
Control tasks for the autonomous excavator can be divided into high-level tasks and low-level tasks. High-level tasks are characterized by complex and computationally intensive tasks whereas low-level tasks are closed-loop control algorithms. Low-level functions require real-time processing with minimal latency. High-level tasks are implemented on powerful pc-based computers. Low-level tasks are run on specialized micro-controllers. Figure 4 illustrates the different components of the system and the way they are connected. Actuators The excavators actuation is realized through a state-of-the-art hydraulic system. However, all actuators are operated directly by the operator. In order to enable digital closed-loop computer control, all actuators have to be fitted with electric interfaces. This is realized through electro-hydraulic pilot valves that work in parallel to the manually operated existing valves.
Processing Unit
Sensors Localisation GPS Inertial
USB CAN
Environment Ethernet Laser range Ethernet ToF Camera Arm Joints Position
Actors
SSI Analog
Pressure
Computer
CAN Microcontroller
PWM
Arm Joints Pilot Valve
Fig. 4. Control system architecture of the excavator
The digging arm itself has 4 DOF, powered by linear hydraulic cylinders. The superstructure is turned through a rotational hydraulic motor. The bucket itself is equipped with a tilting mechanism which adds another DOF. Summing up, there is a total of 6 DOF used in this project (see figure 5): – – – – –
Boom cylinder Adjust cylinder Arm cylinder Swing motor Bucket tilt cylinder
Each one of these hydraulic actuators is controlled by one pilot valve for each direction of movement (FW-BW). Sensors As shown in figure 4, sensors can be categorized in three groups based on their purpose in the system: – Localization Sensors – Environmental Sensors – Kinematic Sensors Localization is realized through GPS sensors and orientation sensors. Information regarding the environment surrounding the excavator is gathered through laser range measuring devices. Time-of-Flight (ToF) camera systems are used to gather detailed information about the perimeter of the digging arm or the filling degree of the bucket. The third category of sensors acquires information about the excavator itself. For closed-loop control of its linear actuators, position feedback is required. This is realized through magnetic restrictive position
Elements: Two-piece boom Boom Superstructure Sensors: valve pressure Rotation angle
Sensors: Cylinder length Cylinder pressure Rotation angle Elements: Dipper arm Bucket
Actor: pilot valves
Fig. 5. Experimental EW180B Volvo bucket excavator
measuring technology. The rotation of the superstructure is acquired through a rotational absolute angle sensor. The control of the pilot valves requires feedback of the control current and the pilot pressure. Each actuator is operated by two pilot valves. Beyond the sensor devices required for closed loop control of the digging kinematics, both hydraulic pressures of each actuator in the system are acquired through pressure transducers. This makes it possible to estimate external forces acting on the bucket, i.e. digging forces.
Data Processing Unit for High-Level Tasks The sensors acquiring information about the environment or position of the excavator produce extensive amounts of data. Sensor data originating from various sensors has to be evaluated and combined to become useful for the automation task. This process needs extensive computing power, which is provided with pc-based computers equipped with powerful cpu’s. The system is modular as lacking computational power can be compensated by adding further computers. These communicate using standard ethernet hardware. Beyond the evaluation of sensor data, the data processing unit serves various further tasks, i.e. – – – – – – – –
Graphical user interface Manual control interfaces (HMI) Conversion of sensory data into an environment model Generation of strategies for the digging task Processing trajectories of the arm Behavior based control algorithms obstacle avoidance Kinematic calculation for each joint motion Security functions
Low-Level Processing Units Closed-loop control requires low-latency processing devices that are equipped with required I/O interfaces and which ensure real-time performance. This is realized through digital signal processors combined with logic programmable devices for the connection of different interfaces. The DSP’s are programmed to meet hard real-time specifications. Transducers required for closed loop control, i.e. pressure and position sensors, are directly connected to the DSP boards. Current control for pilot valves is realized through pulse width modulated amplifiers also directly connected to the DSP’s I/O interfaces.
6
Outlook
The project described in this publication aims to create a first approach towards autonomous excavators. After the realization of the objectives autonomous trenching and landscaping, further works are planned in this field. As the challenges are numerous, the next step will be to incorporate the driving function of the excavator and the interaction with other construction vehicles, i.e. trucks for the transportation of digging material. However, besides the feasibility of an autonomous operation of an excavator the second outcome of the project might be the usage of partial solutions in assistive systems, which could possibly be ported to todays excavators and improve the productivity. The actual behavior-based software control system uses the described algorithms for surface evaluation and localization, based on three dimensional laser scans of the environment, to excavate a trench at a given position. Although a lot of research has to be done and more complex perception algorithms have to be developed, the first steps into the area of autonomous mobile excavators have been successfully taken.
References 1. D. A. Bradley and D. W. Seward, “The development, control and operation of an autonomous robotic excavator,” Journal of Intelligent and Robotic Systems, vol. 21, no. 1, pp. 73–97, November 2 2004. 2. Y. Sakaida, D. Chugo, K. Kawabata, H. Kaetsu, and H. Asama, “The analysis of excavator operation by skillful operator,” in Proc. of 23rd International Symposium on Automation and Robotics in Construction, 2006, pp. 543–547. 3. H. Cannon, “Extended earthmoving with an autonomous excavator,” Master’s thesis, Carnegie Mellon Robotics Instiute, 1999. 4. P. Rowe and A. Stentz, “Parameterized scripts for motion planning,” in IEEE/RSJ International Conference on Intelligent Robots and Systems, IROS 97, vol. 2, 1997. 5. A. Stentz, J. Bares, S. Singh, and P. Rowe, “A robotic excavator for autonomous truck loading,” Autonomous Robots, vol. 7, no. 2, pp. 175–186, September 1999. 6. R. Arkin, Behaviour-Based Robotics. MIT Press, 1998. 7. M. Proetzsch, T. Luksch, and K. Berns, “Development of complex robotic systems using the behavior-based control architecture iB2C,” Robotics and Autonomous Systems, vol. 58, no. 1, 2010.
