Multi Agent Systems Learn to Safely Move Indoor Environment

This letter presents a path-planning algorithm for a fleet of autonomous agents operating in a bounded indoor environment with static and moving obstacles. The proposed algorithm uses a combination of Modified Artificial Potential Field (MAPF) and Reinforcement Learning (RL) to determine safe paths for the agents to their respective goal locations. The proposed approach ensures avoiding collision with the obstacles and among the agents. The better performance of our proposed method, suitable for a real-world operation, is illustrated by comparing it with multiple RL and MAPF concepts. In addition to simulations, we carry out practical experiments using multiple open-source flying development platforms in an indoor VICON lab environment to demonstrate the efficacy of the proposed approach.