Reinforcement learning-based optimization of locomotion controller using multiple coupled CPG oscillators for...

Area of Science:

• Robotics
• Biomimetics
• Control Systems

Background:

• Robotic locomotion often struggles to replicate the natural, fluid movements of aquatic animals.
• Existing controllers may lack the adaptability for complex, multi-parameter swimming patterns.

Purpose of the Study:

• To develop an adaptive locomotion controller for undulating elongated fin robots inspired by black knifefish.
• To enable natural swimming pattern transformations and parameter control (amplitude envelope, oscillatory frequency).

Main Methods:

• A modified Central Pattern Generator (CPG) network with sixteen coupled Hopf oscillators was designed.
• Reinforcement learning was employed to optimize the convergence rate of the modified CPG network.
• The controller integrates fin-ray angle feedback for enhanced control.

Main Results:

• The proposed controller successfully enabled natural swimming pattern transformations in the robot.
• Various swimming patterns were achieved by configuring amplitude envelope and oscillatory frequency parameters.
• Simulation and experimental results validated the controller's capability and effectiveness across different parameters.

Conclusions:

• The black knifefish-inspired locomotion controller offers a robust and adaptable solution for undulating fin robots.
• Reinforcement learning optimization significantly enhances the CPG network's performance and control capabilities.
• This approach paves the way for more biomimetic and versatile underwater robotic systems.