OBSTACLE-BASED ROUTE PLANNING FOR A STEERING CONSTRAINED AUTONOMOUS VEHICLE

The Team Cybernet vehicle for the 2007 DARPA Urban Challenge 1 incorporated a route planning approach that uses sensed obstacles in the environment as the basis for potential turn placement prior to performing path search. The path search is confined to finding a set of straight-line tangents that connect circles of maximum curvature that are constructed adjacent to sensed obstacles. This approach is substantially different from traditional approaches in that the complexity of the search space is not based on the length of the path, but rather on the number of obstacles in the field. For sparse obstacle fields, this approach allows for very fast plan generation and results in paths that are guaranteed by construction to not violate steering constraints. 1 DISCLAIMER: The information contained in this paper does not represent the official policies, either expressed or implied, of the Defense Advanced Research Projects Agency (DARPA) or the Department of Defense. DARPA does not guarantee the accuracy or reliability of the information in this paper. Proceedings of the 2011 Ground Vehicle Systems Engineering and Technology Symposium (GVSETS) Obstacle-Based Route Planning for a Steering Constrained Autonomous Vehicle, Rowe et al. Page 2 of 7 Introduction In the 2007 DARPA Urban Challenge, one of the qualification criteria was the ability for the autonomous vehicle to traverse bounded, open “zones” containing obstacles and parking spaces. A typical zone mission involved entering the zone from a roadway, navigating to a parking space within the zone, pulling into a parking space, backing out of the parking space, and then proceeding to a defined zone exit point (see Figure 1). Figure 1: A Typical Zone Mission. The dotted line represents a potential vehicle path, where each dot represents one node in the search tree. The circles and the lines that link them show a path generated with our approach. Our over-arching vehicle design philosophy was to solve exactly the problems we had to, and not go any further. Scoring for this event was based on avoiding obstacles (including lines on parking spaces), providing a cushion of at least 1 meter around all other vehicles, reaching the destination checkpoints, and not pausing for more than 5 seconds 2 . This guided our choice of solution. We made certain assumptions about the context in which the planning was to take place. We assumed that the area would 2 The complete scoring documentation for the 2007 Urban Challenge can be found at http://www.darpa.mil/GRANDCHALLENGE/rules.asp. be typical of a “real” parking lot, i.e. large (100x50 meters was a working baseline), flat (no obstacles obscured by the ground), and most importantly that the nature of the obstacles would be islands that needed to be driven around, rather than forming a maze that had to be driven through. We also assumed that DARPA might add a cul-de-sac to the zone to test the vehicle’s ability to plan out of local minima. Figure 2: Team Cybernet Urban Challenge Vehicle This motivated the path planning algorithm to be described in this paper. The algorithm constrains the planning process to find paths that achieve the minimum obstacle avoidance distance and that do not require turns smaller than the vehicle’s minimum turning radius to follow. The two constraints can be used to generate a reduced planning graph space to be searched as compared to more traditional discrete grid quantization of the space, and because the space is not quantized, allow any possible smooth path that meets the constraints to be generated.