Preprocessing Configuration Space for Improved Sampling Based Path Planning Titas Bera a,1 , M. Seetharama Bhat a , D. Ghose a a Department of Aerospace Engineering, Indian Institute of Science, Bangalore Abstract: Sampling based planners have been successful in path planning of robots with many degrees of freedom, but still remains ineffective when the configuration space has a narrow passage. This paper presents two new techniques of preprocessing the configuration space. The first technique called a Random Walk to Surface (RWS), uses a random walk strategy to generate samples in narrow regions quickly, thus improving efficiency of Prob- abilistic Roadmap (PRM) based planners. The algorithm substantially reduces instances of collision checking and thereby decreases computational time. The method is powerful even for cases where the structure of the narrow passage is not known a priori, thus giving sig- nificant improvement over other known methods. The second method, by preprocessing the configuration space, improves the efficiency of Rapidly Exploring Random Tree (RRT) like planners by identifying key regions of the configuration space to search for a solution path. The Experiments show a significant improvement in efficiency for both PRM and RRT like planners. Key words: Robot motion planning, randomized algorithm, PRM, RRT 1. INTRODUCTION Several areas beyond classical robotics, are benefiting from advances in algorithmic motion planning. Examples include structural studies in biology, computer games, path planning of unmanned aerial vehicles among the others. The success of these algorithms on different instances is because of availability of an abstract formulation of the model of many different real world applications, and ability to solve difficult problems in a reasonable time. Modern motion plan- ning begins with the introduction of the notion of configura- tion space by Lozano-Perez and Wesley [1]. In the configu- ration space the robot is reduced to a point; hence the motion planning problem becomes that of finding a path from an ini- tial point to a goal point in the configuration space. However, the explicit mapping from workspace obstacles to configura- tion space is in general difficult. Early studies showed that the basic version of this problem is PSPACE-complete [3], [5] and the best exact deterministic algorithm known is ex- ponential in the dimension of the configuration space. On the other hand, real world problems generate instances with high dimensional configuration spaces. Email addresses: titasbera@aero.iisc.ernet.in (Titas Bera ), msbdcl@aero.iisc.ernet.in (M. Seetharama Bhat), dghose@aero.iisc.ernet.in (D. Ghose) URL: http://www.aero.iisc.ernet.in (Titas Bera ) 1 Corresponding author Since the mid nineties, in order to break this ”curse of di- mensionality”, sampling based approaches were introduced. The approach is based on the generation of samples to ac- quire information about the problem instance being solved. The first generation of these algorithms heavily relied on ran- dom samples. In contrast, there has been a recent trend to use deterministic sampling schemas [20]. For motion plan- ning problems, samples are stored in a data structure which represents an approximation of the configuration space, as opposed to its exact combinatorial representation. The data structure is usually composed of nodes, that is, samples in the configuration space, and links, that is, valid paths connecting samples. Nodes and links can be stored in the form of graphs or trees. The entire continuum of configuration space is then approximated to a network of nodes. The implementation of these algorithms is usually quite simple. The price to pay is completeness. Traditional combinatorial motion planning algorithms are complete, that is, they will find a solution if one exists, and will report failure otherwise. Algorithms based on randomly generated samples aims for probabilistic completeness. This means that if a solution exists, the probability to find it con- verges to 1 when the computation time approaches infinity [6]. Despite probabilistic completeness, randomized algo- rithms such as the Randomized Potential Field Planner (RPP) [7], the Probabilistic Road-map (PRM) family [8], Rapidly- Proceedings of ICEAE 2009