RA-RRTV*: Risk-Averse RRT* With Local Vine Expansion for Path Planning in Narrow Passages Under Localization Uncertainty

Abstract

Recent advances in sampling-based algorithms have enhanced the ability of mobile robots to navigate safely in environments with localization uncertainty. However, navigating narrow passages remains a significant challenge due to the heightened risks posed by uncertainty. In this letter, we present a novel algorithm, Risk-Averse RRT* with Local Vine Expansion Behavior (RA-RRTV*), to systematically address these challenges. The algorithm combines RRT* with chance constraints and incorporates an objective function to balance path length and risk, enabling the discovery of risk-averse paths. Narrow passages in the belief space are identified using sample-based information, while sequential Bayesian sampling is employed to guide the expansion of local belief vines, ensuring connectivity in high-risk regions. We provide proof of the asymptotic optimality of RA-RRTV*. The effectiveness of RA-RRTV* is demonstrated through extensive simulations and real-world experiments.