This academic article discusses the development of a decentralized collision avoidance policy for Autonomous Surface Vehicles (ASVs). The policy is based on Distributional Reinforcement Learning (DRL), which considers the interactions among ASVs as well as with static obstacles and current flows. The policy was evaluated against a traditional RL-based policy and two classical methods, showing superior performance in navigation safety while requiring minimal travel time and energy. The article also introduces a variant of the framework that automatically adapts its risk sensitivity to improve ASV safety in highly congested environments.
Publication date: 20 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.11799
