Probabilistic model based path planning

Abstract

Path planning shows great potential for exploring indoor and outdoor environments. In this paper, a probabilistic method is proposed to design path planners based on transition probabilistic matrices and signed distance functions. The transition probabilistic matrix is constructed by collecting path sequence data generated by performing a modified RRT with many times. Moreover, the signed distance function is introduced to simulate the safety coefficient which can guarantee a suitable distance between robots and obstacles. By combining the transition probability and the safety coefficient, our path planning task is modeled as a maximal probability sequence decision problem which in essence is equivalent to a minimal cost path problem, and then the dynamic programming solver is achieved by using the push-based efficient implementation of Bellman–Ford’s algorithm Kleinberg and Tardos (2006). Several path evaluation criteria are also used to evaluate path planning results, and plenty of experimental results illustrate the effectiveness of the proposed method.