On the development of autonomous car-like vehicles.

Gall, Robert ; Troester, Fritz ; Luca, Razvan 等

1. INTRODUCTION

Given the need for developing new intelligent car-like mobile robots and thanks to a previous experience in this field, the Heilbronn University (Germany) has established a research center for fully autonomous parking in collaboration with the company Valeo (Bietigheim-Bissingen, Germany) and the Transilvania University (Brasov, Romania). The aim of our project is to develop (design, simulate and implement) a system that is able to accomplish specific tasks that allow a driverless car to fully autonomously execute, for example parking procedures on standard parking lots. Considering the navigation of autonomous mobile systems in partially known or unknown dynamic environments, the complete contact-less sensory coverage of the workspace represents a fundamental difficulty. In this sense in the following we present our approach and solutions to the problematic of controlling an autonomous nonholonomic car-like mobile robot focusing on the simultaneous localization and mapping (SLAM), the sensor based path planning and on the implementation of the control algorithms on a real-time Matlab xPC based computation machine.

Because autonomous vehicles are in fact specifically designed mobile robots, developing autonomy levels for these systems is a complex issue that has to take into account many factors such as task complexity, human interaction, environmental difficulty, system dependence, and quality factors. Driver assistance has become more and more performant so that semiautonomous driving can be nowadays achieved.

The delimitation of the environment includes the following scenario: on a randomly configured parking lot with a previously unknown number of parked vehicles, a virtual car needs to drive trough according to the pre-planned path and identify empty parking spots.

Following this idea we have created a strategy that eases up the developing process of the systems. Basically we work in parallel on 3 levels: simulation of real time traffic scenarios (Matlab Simulink), implementation of algorithms on a scaled model vehicle (Robocar) and finally the testing of the programs on the real automobile (A Class Mercedes).

2. RELATED WORK

In order to understand the motivation for the development of autonomous car-like mobile systems we analyze related work. Due to the possibilities based on technical progress on one hand and the increasing demand for active and passive safety on the other hand, the quantity and quality of accessory systems for cars increased significantly over time (Kroedel, 2008). Some research centers, like the EPFL's Autonomous Systems Lab (Lamon, 2006), have designed a system that can show fully autonomous navigation and 3D mapping in outdoor settings. Their full scaled mobile robot, constructed on the architecture of a standard Smart vehicle uses five distance laser sensors, three cameras, a differential GPS and an Inertial Measurement Unit (IMU) and four computers.

Another interesting development is the MERLIN, developed by the Univ. of Wurzburg (Zeiger, 2006) and presented at the European Land Robotics Trial, a scaled 1:8 rover with semi-autonomous functions to handle rough terrain. The mobile robot can be easily used as a prototype for the implementation of SLAM (Simultaneous Localization and Mapping) and path planning algorithms.

In the field of path planning and obstacle avoidance two solutions are proposed in (Brath, 2005) and (Hrich, 2005). The first is based on the mathematical theory of differential games. The second one, called "the elastic bands", was developed by Quinlan and Khatib and relies on repulsive potential field generation. More precisely, the road (the left and right side) and the obstacles are modeled on an elastic network which is composed of several nodes linked up with springs. The spring's stiffness is related to the desired path, which is the equilibrium position of the network in the absence of the obstacles (Pozna, 2005).

3. PARKING SCENE SIMULATION

One of the first issues we have been facing since we started working on the simulation was that we had to define the models and the model specific interfaces. The software programming language being set, MATLAB Simulink, some basic strategically set up blocks have been defined.

The central computer system receives information from the cars sensors based on witch navigational decisions are made. At the same time a map is build and the whole scene is interpreted.

As a target, the identification of free parking spaces and the recognition and avoidance of obstacles are crucial. The virtual scene is shown in fig.1. The computed path (1) of the vehicle and the real followed trajectory are highlighted as well as the results delivered by the laser scanner (2) (marked with small circles).

[FIGURE 1 OMITTED]

The dynamic model and gathered odometry information's are used for the transformation of measurement data from local to global coordinates. The scene's random generation is applied to parked vehicles and to the noise added to laser measurements. The SLAM problem is solved by means of the ICP (Iterative Closest Point) and DCE (Discrete Contour Evolution) algorithms. Also we have developed a parking specific simulation with 3 parallel and 2 transversal parking strategies (fig. 2). The system works as following:

* The "driver" activates the system by pushing a button.

* The system measures the size of the parking slot and determines whether the car fits into it or not.

* Once a slot has been found the system stops and puts the car into reverse thereby activating specific ultrasonic sensors and object avoidance modules. Semi- and fully autonomous parking is achieved.

[FIGURE 2 OMITTED]

4. PATH PLANING AND HARDWARE

The trajectory is generated by means of 3rd degree Bezier curves while the controller (PID and State-Space) features a newly proposed look-ahead algorithm witch predicts the followed path and adapts accordingly. We analyze the upcoming 2 to 10 m of the path, in dependency of the cars velocity so that the controller can easelier make the adjustments. A major improvement has been noticed in the minimization of the path-following errors.

[FIGURE 3 OMITTED]

As an intermediate step between the simulation and the autonomous car a scaled mobile car-like robot (fig. 4) has been developed. On the existing mechanic we integrated an xPC (Matworks, 2009), a sensorial system that consists of one laser scanner, 2 clusters of 10 ultrasonic sensors and sensors that monitor odometry data. The target xPC wirelessly communicates with a host laptop on which an application specific GUI is running. Processes that need real-time computations efficiency are downloaded on the target. A PWM signal controls the vehicles acceleration and direction. The third step was to integrate a robotized module into an A Class Mercedes to allow autonomous driving and navigation.

[FIGURE 4 OMITTED]

First attempts were made with the ACC Fahrautomat 1 (Troester, 2004), then we developed a second generation, optimized variant of the system, presented in (fig. 3) (Gall, 2008).

For a better positioning we have added a DGPS unit (static and mobile platform), delivered by the company GeneSys, which allows us to perform measurements with up to 20 mm precision (x, y).

5. CONCLUSIONS AND FUTURE WORK

This paper presents only partial results of the research. The aim was to present a strategy for designing autonomous systems that can perform a fully automated parking in real traffic conditions. At the actual state only part of the parking procedure can be done on different levels (simulation, model vehicle and real automobile). This means that the real robotized car can perform separately the following procedures: Follow a predefined trajectory in fully autonomous mode; Identify a parking spot and make semiautonomous parking; Scan the environment and identify the contour of the surrounding by means of laser, ultrasonic, radar and video sensors. For the near future we are planning to improve the SLAM, integrate obstacle avoidance into the path planning strategies and fuse all of these modules into the system.

6. REFERENCES

Brandt, T., Sattel, T. Wallaschek, J. (2005); On automatic Collision Avoidance Systems. SAE, Budapest,Hun, 2005.

Gall R., Troester F., Mogan Gh. (2008); Research upon the development in the field of Autonomous vehicles, Proceedings of the 4th International Conference Robotics'08, (vol. II, pp 673-678), Brasov, 2008

Hrich, K. (2005); Optimization of Emergency Trajectories of Autonomous Vehicles with Respect to Linear Vehicles Dynamic, IEEE/ASME 2005 International Conference on Advance Intelligent Mechatronics California SUA 2005.

Krodel M. (2008); Towards a Learning Autonomous Driver System, Available from: http://www.netcologne.de, Accessed: 2009-01-20

Lamon P., Kolski S., Triebel R., Siegwart R., Burgard W. (2006); The SmartTer for ELROB 2006--A Vehicle for Fully Autonomous Navigation and Mapping in Outdoor Environments, Available from: http://www.smart-team.ch, Accessed: 2009-01-20

Pozna C., Troester F. (2005); The inverse cinematic of the ACC mobile robot, Proceedings of the ninth IFTOMM international symposium on theory of machines and mechanisms, (vol. I, pp 358-366) Bucharest, 2005

Troester F., Pozna C. (2004); The ACC Fahrautomat Project, Proceedings of the International Conference Robotica, Timisoara 2004

Zeiger F., Schilling K. (2008); Design of an User Interface for the Coordination of a Group of Mobile Robots, The 17th International Symposium on Robot and Human Interactive Communication, Munich (Germany), 2008