Swimming strategy of settling elongated micro-swimmers by reinforcement learning

Jingran Qiu, Lihao Zhao, Chunxiao Xu, Yichen Yao

Published 2018-11-27Version 1

Motile microorganisms in aquatic ecosystems are able to sense the surrounding environment and adjust their motion to reach certain regions that are favourable for their growth or reproduction. Studying the moving strategies of microorganisms is important for an in-depth understanding of their behaviour in aquatic environment. In present work, we model microorganisms as smart swimming particles and introduce reinforcement learning to investigate the strategy for moving upward in a two-dimensional flow field. We explore how gravity and elongation of a particle affect the strategies obtained by reinforcement learning and compared with naive gyrotactic particles. We examine the micro-swimmers with different motilities (quick-alignment and slow-alignment). Under the same conditions of quick-alignment motility and flow configuration multi-solutions of swimming strategy are observed in the case of smart particles trained by reinforcement learning. However, the multi-solutions are converged into an almost optimal strategy with inclusion of gravity, which acts as a constraint on particle motion. Moreover, the elongation of particle is found to enhance the ability of particle in sampling the low vorticity and upwelling region. When the settling and elongation are both considered for slow-alignment particles, similar performance of moving upward is observed for both smart and naive particles. The interesting findings indicate that the diversity of proper strategies is restricted with including constraints of more realistic factors and we suspect that the elongation and gyotaxis of microorganism might be an almost optimal strategy for swimming upwards after the long term natural selection. Additionally, the current work on the swimming strategies of more realistic particles reveals the effectiveness of reinforcement learning in the study of the behaviour of microorganisms in fluid flow.