Decentralized decision-making mechanisms for efficient locomotion in deformable microswimmers

Determine how decentralized decision-making in a deformable microswimmer leads to efficient collective locomotion of its body parts, including identifying the local perception-action processes and coordination mechanisms by which individually controlled body segments produce global propulsion under force-free, low-Reynolds-number conditions.

Background

Most prior reinforcement learning studies on microswimmers model them as rigid agents with centralized control, limiting scalability and ignoring internal degrees of freedom. Biological microswimmers, however, achieve locomotion via local, distributed actuation (e.g., cilia or molecular motors) without a centralized controller, suggesting that efficient navigation may emerge from decentralized coordination. Despite these observations, the precise mechanisms by which local decisions and interactions between body parts yield efficient global swimming remain insufficiently characterized.

This paper investigates decentralized control by equipping each bead in a generalized Najafi–Golestanian N-bead swimmer with identical artificial neural network controllers that perceive only local information and act independently. The paper demonstrates that efficient locomotion can emerge from such decentralized policies and that these policies generalize across morphologies, but the broader mechanistic understanding of how decentralized decisions generate optimal collective gaits is framed as an explicit open question.