Smart Magnetic Microrobots Learn to Swim with Deep Reinforcement Learning.

Area of Science:

• Robotics
• Biomimetic Engineering
• Artificial Intelligence

Background:

• Microrobots with complex designs require robust control in unpredictable environments.
• Modeling system dynamics and achieving precise positional control for microrobots is challenging.

Purpose of the Study:

• To develop a smart microrobot capable of autonomous navigation in uncharacterized environments.
• To utilize deep reinforcement learning for deriving adaptive control policies without system modeling.

Main Methods:

• Development of a helical magnetic hydrogel microrobot.
• Application of the soft actor critic reinforcement learning algorithm.
• Control via a time-varying magnetic field from a three-axis electromagnet array.

Main Results:

• The reinforcement learning agent successfully learned control policies from state vectors and raw images.
• The learned policies enabled the microrobot to swim through a biomimetic fluidic environment.
• The autonomous control strategies mimicked those of controllers based on physical models.

Conclusions:

• Deep reinforcement learning offers a powerful approach for creating adaptive microrobot controllers.
• This method significantly enhances microrobot capabilities for operation in complex, uncharacterized environments.
• Future microrobot generations will benefit from AI-driven control strategies.