Knee exoskeleton enhanced with artificial intelligence to provide assistance-as-needed

Area of Science:

• Robotics
• Artificial Intelligence
• Rehabilitation Medicine

Background:

• Robotic therapy is crucial for motor function recovery in stroke patients.
• Current robotic training often uses predetermined trajectories, limiting adaptability.
• Active patient participation is key, necessitating adaptive assistance methods.

Purpose of the Study:

• To introduce AI agents for autonomous control of knee exoskeletons in rehabilitation.
• To propose and evaluate a novel assist-as-needed (AAN) cooperative control method.
• To investigate the efficacy of AI-assisted human-robot cooperative control.

Main Methods:

• Development of an electromyographic (EMG)-controlled knee exoskeleton integrated with a screen game.
• Implementation of two AI agents (modular pipeline, deep Q-network) for exoskeleton control.
• Comparison of AAN strategy against a fixed assistant ratio in human-robot cooperative control.

Main Results:

• AI agents achieved higher game scores than healthy subjects using EMG control alone.
• The AAN method outperformed the fixed assistant ratio, evidenced by higher muscle activation and participant scores.
• Human-robot cooperative control significantly increased integral muscle activation, indicating greater subject involvement.

Conclusions:

• AI agents show potential for assisting stroke rehabilitation through exoskeleton control.
• The AAN cooperative control strategy enhances patient engagement and rehabilitation outcomes.
• AI-powered rehabilitation robots offer a promising avenue for personalized and adaptive therapy.