Self-learning how to swim at low Reynolds number

Designing locomotory gaits for synthetic microswimmers has been a challenge due to stringent constraints on self-propulsion at low Reynolds numbers (Re). Here, we introduce a new theoretical approach of designing a class of self-learning, adaptive (or "smart") microswimmers via reinforcement learning. Diverging from the traditional paradigm of specifying locomotory gaits a priori, here a self-learning swimmer can develop and adapt its propulsion strategy based on its interactions with the surrounding medium. We illustrate this new approach using a minimal but representative model swimmer consisting of an assembly of spheres connected by extensible rods. Without requiring any prior knowledge of low Re locomotion, we demonstrate that this self-learning swimmer can recover a previously known propulsion strategy, identify more effective locomotory gaits, and adapt its locomotory gaits in different media. This approach opens an alternative avenue to designing the next generation of smart microrobots with robust locomotive capabilities.