Multi-robot path planning in factory-like environments

Due to the growing popularity of systems with multiple cooperative robots, there is a growing need to solve the problem of multi-agent path planning: determining how a group of robots should optimally reach their respective goals with minimal inter-robot interference. Previous work done in this eld takes too long to plan for real-time systems, especially for complex, obstacle-ridden environments. In an attempt to remedy these problems we propose a new method of multi-agent path planning using distributed deep reinforcement learning where robots collectively learn a single policy for path planning which can then be applied to each robot individually. This approach would advance the state of the art by providing a fast, scalable, and robust method of decentralized multi-agent path planning.

People: Guillaume Sartoretti, Lenny Zhang, Denise Li, Josh Durham,Justin Kerr