Reinforcement learning of a biflagellate model microswimmer

Area of Science:

• Physics
• Biophysics
• Fluid Dynamics

Background:

• Microswimmers utilize periodic, non-reciprocal appendage deformations for locomotion in viscous fluids.
• Biflagellate models, inspired by biological swimmers, often use simplified flagellar motions.

Purpose of the Study:

• To identify quasi-optimized swimming strokes for a simple biflagellate microswimmer model using reinforcement learning.
• To compare the performance of identified strokes against existing models.

Main Methods:

• A computational model of a microswimmer with a spherical body and two flagellar beads was developed.
• Reinforcement learning algorithms were employed to discover effective swimming gaits.
• Flow fields generated by the microswimmer were analyzed.

Main Results:

• Reinforcement learning identified quasi-optimized, symmetric, and quasi-synchronized beating patterns for the flagella.
• The identified gaits resulted in pusher-type flow fields.
• The novel swimming gaits demonstrated superior performance compared to predefined circular flagellar motions.

Conclusions:

• Reinforcement learning is effective in discovering efficient microswimming strategies.
• Optimized flagellar coordination significantly enhances microswimmer performance.
• The pusher-type flow field is a characteristic of these efficient gaits.