Path Set Relaxation for Mobile Robot Navigation

Abstract

This paper addresses autonomous navigation and goal acquisition for mobile robots operating in difficult, cluttered environments. In particular a hierarchical approach to navigation is of interest, which subdivides the problem into global and local components. Local planners attempt to search the continuum of actions for a best (safest, efficient) route towards a goal. To achieve real-time performance, the search space is often sampled in lowdimensional action or state space. This paper explores a relaxation-based approach that optimizes the sampled action space for the perceived environment. The gradientbased approach minimizes the cost of each motion in the path set until convergence into a local optimum is reached. Simulation experiments show that relaxed arc sets offer better approximations of the acceptable path continuum and lead to safer navigation in rough terrain and dense obstacle fields. Additional experiments explore the benefits of sampled action spaces composed of higher-order action primitives (clothoids) and a graduated fidelity inspired lookahead technique.