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

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