Reinforcement learning for pursuit and evasion of microswimmers at low Reynolds number

Abstract

Using reinforcement learning, we study the coevolution of pursuing-evasion policies of two microswimmers that can sense each other only through hydrodynamic signals, which provide ambiguous information. We show that both agents find effective ways to overcome the difficulties set by partial information, and we explain the main discovered strategies. The setting here developed may offer a framework to study prey-predator interactions in more complex situations.

Topics & Concepts