Reinforcement learning of a biflagellate model microswimmer

TL;DR

This paper employs reinforcement learning to optimize swimming strokes in a simple biflagellate microswimmer model, revealing symmetric flagella beating and flow fields that outperform traditional models.

Contribution

It introduces a reinforcement learning approach to discover efficient swimming gaits for a biflagellate microswimmer model, surpassing previous predefined motion strategies.

Findings

Identified quasi-optimized, symmetric flagella beating patterns.

Flow fields resemble pusher-type microswimmers.

Outperforms traditional predefined motion models.

Abstract

Many microswimmers are able to swim through viscous fluids by employing periodic non-reciprocal deformations of their appendages. Here we use a simple microswimmer model inspired by swimming biflagellates which consists of a spherical cell body and two small spherical beads representing the motion of the two flagella. Using reinforcement learning we identify for different microswimmer morphologies quasi-optimized swimming strokes. For all studied cases the identified strokes result in symmetric and quasi-synchronized beating of the two flagella beads. Interestingly, the stroke-averaged flow fields are of pusher type, and the observed swimming gaits outperform previously used biflagellate microswimmer models relying on predefined circular flagella bead motion.