Energy-Aware Multi-Goal Motion Planning Guided by Monte Carlo Search

Abstract

Autonomous robots need a reliable way to preserve their energy level while performing a persistent task such as inspection or surveillance. Toward this objective, this paper considers the multi-goal motion-planning problem with multiple recharging stations where a robot operating in a complex environment has to reach each goal while reducing the travel distance and the number of times it recharges. This paper develops an integrated approach that couples sampling-based motion planning with Monte-Carlo Tree Search (MCTS). The proposed MCTS searches over a discrete abstraction, which is obtained via a probabilistic roadmap, and uses a reward function to calculate when, where, and whether it is beneficial to recharge. This results in short tours that also reduce the number of recharges. Such tours are used to guide sampling-based motion planning as it expands a tree of collision-free and dynamically-feasible motions. Experiments with nonlinear dynamical robot models operating in obstacle-rich environments demonstrate the efficiency of the approach.