Continuous mode adaptation for cable-driven rehabilitation robot using reinforcement learning.

Area of Science:

• Robotics
• Rehabilitation Engineering
• Control Systems

Background:

• Continuous mode adaptation is crucial for rehabilitation robots to meet diverse user needs and enhance human-robot interaction (HRI).
• Existing methods often require detailed human and robot dynamics models, limiting adaptability.
• Cable-driven rehabilitation robots (CDRR) offer unique advantages but require sophisticated control strategies for effective rehabilitation.

Purpose of the Study:

• To propose a reinforcement-learning-based optimal admittance control (RLOAC) strategy for CDRRs.
• To enable continuous mode adaptation between passive and active working modes without prior knowledge of dynamics models.
• To improve HRI performance and user rehabilitation outcomes.

Main Methods:

• A reinforcement learning algorithm was employed to determine optimal admittance parameters by minimizing a cost function.
• The cost function incorporated trajectory error and human voluntary force.
• Contribution weights within the cost function were modulated based on human voluntary force for adaptive control.

Main Results:

• Simulations and experiments with 10 subjects validated the RLOAC strategy's feasibility and effectiveness.
• The RLOAC strategy demonstrated notably lower tracking error and energy consumption per unit distance compared to traditional admittance control.
• Experimental results confirmed improved tracking accuracy and robot compliance.

Conclusions:

• The proposed RLOAC strategy effectively achieves continuous mode adaptation in CDRRs.
• RLOAC enhances rehabilitation robot performance by improving accuracy and compliance.
• This strategy shows significant potential for future applications in rehabilitation robotics.