The Emergence of Complex Bodyguard Behavior Through Multi-Agent Reinforcement Learning

In this paper we are considering a scenario where a team of robot bodyguards are providing physical protection to a VIP in a crowded public space. We show that the problem involves a complex mesh of interactions between the VIP and the robots, between the robots themselves and the robots and the bystanders respectively. We show how recently proposed multi-agent policy gradient reinforcement learning algorithms such as MADDPG can be successfully adapted to learn collaborative robot behaviors that provide protection to the VIP.