Motion Intention Recognition and DDPG-Based Adaptive Impedance Control for a Robotic Upper-Limb Exoskeleton.

Area of Science:

• Robotics
• Biomedical Engineering
• Artificial Intelligence

Background:

• Upper-limb motor impairments significantly impact daily living.
• Current rehabilitation methods may lack adaptability for diverse patient needs.
• Robotic exoskeletons offer potential for enhanced, personalized rehabilitation.

Purpose of the Study:

• To present a metamorphic robotic exoskeleton for upper-limb rehabilitation.
• To develop and evaluate an AI-driven framework for configuration recognition and trajectory prediction.
• To implement an adaptive impedance controller for compliant human-robot interaction.

Main Methods:

• A metamorphic mechanical architecture with four configurations (shoulder and elbow movements, forearm rotation).
• Termite Life Cycle Optimizer-tuned Support Vector Machine (TLCO-SVM) for configuration recognition.
• TLCO-optimized Long Short-Term Memory (TLCO-LSTM) network for joint angle prediction.
• Deep Deterministic Policy Gradient-based adaptive impedance controller for assistive torque generation.

Main Results:

• TLCO-SVM achieved 98.10% average classification accuracy.
• TLCO-LSTM demonstrated low Root Mean Square Errors (RMSEs) for joint angle prediction (e.g., 2.41° for SF/E).
• Assistive-torque tracking RMSEs were below 0.35 Nm across all configurations.

Conclusions:

• The proposed AI-driven robotic exoskeleton effectively recognizes configurations and predicts desired movements.
• The adaptive impedance controller ensures safe and compliant physical interaction.
• This system shows promise for advanced upper-limb rehabilitation.