Decentralized Motion Planning for Multi-Robot Navigation using Deep Reinforcement Learning

Abstract

This work presents a decentralized motion planning framework for addressing\nthe task of multi-robot navigation using deep reinforcement learning. A custom\nsimulator was developed in order to experimentally investigate the navigation\nproblem of 4 cooperative non-holonomic robots sharing limited state information\nwith each other in 3 different settings. The notion of decentralized motion\nplanning with common and shared policy learning was adopted, which allowed\nrobust training and testing of this approach in a stochastic environment since\nthe agents were mutually independent and exhibited asynchronous motion\nbehavior. The task was further aggravated by providing the agents with a sparse\nobservation space and requiring them to generate continuous action commands so\nas to efficiently, yet safely navigate to their respective goal locations,\nwhile avoiding collisions with other dynamic peers and static obstacles at all\ntimes. The experimental results are reported in terms of quantitative measures\nand qualitative remarks for both training and deployment phases.\n