Cooperative Emergent Swarming Through Deep Reinforcement Learning
Tony X. Lin, Daniel M. Lofaro, Donald Sofge · 2020
This paper studies the problem of designing a decentralized controller that is able to induce a desired emergent behavior in robot swarms. We consider a robot swarm composing of agents that have identical dynamics and sensors and are unable to communicate among each other. Our proposed approach leverages deep reinforcement learning methods to search for an appropriate control policy that achieves our desired group behavior. We show that our method is able to capture targets in a predator-prey case-study. Simulation and experimental results using an autonomous blimp research platform are also provided.