Reinforcement learning selects multimodal locomotion strategies for bioinspired microswimmers

Area of Science:

• Biomimetic robotics
• Artificial intelligence in fluid dynamics
• Microscale locomotion

Background:

• Natural microswimmers use complex, multimodal strategies for navigation and survival.
• Replicating these sophisticated locomotion behaviors in artificial microswimmers remains a significant challenge.
• Hydrodynamic coupling complicates the direct relationship between actuation and emergent motion.

Purpose of the Study:

• To develop a reinforcement learning (RL) framework for bioinspired microswimmers.
• To enable artificial microswimmers to select task-specific locomotion strategies.
• To demonstrate adaptive navigation capabilities in a model microswimmer.

Main Methods:

• Utilized a reinforcement learning (RL) approach to train a bioinspired microswimmer model.
• Modeled the microswimmer based on *Chlamydomonas reinhardtii*, featuring a sphere body and two flagella spheres.
• Defined learning objectives to optimize for displacement maximization or energy minimization.

Main Results:

• The RL-powered microswimmer successfully selected locomotion strategies tailored to specific learning goals.
• Demonstrated the ability to maximize forward displacement and minimize energy consumption.
• Showcased multi-directional navigation through coordinated switching between forward and steering gaits.

Conclusions:

• Reinforcement learning provides a viable method for endowing bioinspired microswimmers with adaptive multimodal locomotion.
• This approach facilitates the design of artificial microswimmers capable of versatile navigation and task execution.
• Opens new possibilities for advanced bioinspired robotic systems.