Distributed Multi-robot Navigation in Dynamic Environment via Deep Reinforcement Learning

The uncertainty of obstacles in dynamic environments, the local nature of each robot’s observation, and the dynamic interaction between autonomous robots make autonomous navigation of multi-robot systems extremely challenging. To address the above problems, this paper proposes a distributed multi-robot navigation approach based on deep reinforcement learning, which models the multi-robot navigation problem as a partially observable Markov decision process, simulates parametric navigation and obstacle avoidance policies using deep neural networks, and updates the policies through the proximal policy optimization algorithm based on the actor-critic model. A deep reinforcement learning experimental platform based on ROS is built for autonomous learning of optimal navigation policy. The experiment results show that our approach narrows the performance gap between simulation and reality of deep reinforcement learning based navigation and obstacle avoidance policies, and provides a technical reference for autonomous navigation of multi-robot systems in dynamic environments without global precise maps and communication.