Abstract
This study presents a novel AI-driven framework for optimizing upper limb rehabilitation tasks by integrating electromyography (EMG), force-position analysis, and reinforcement learning (RL). The approach combines supervised deep learning models to predict muscle activation levels with a Proximal Policy Optimization (PPO) agent that generates muscle-targeted rehabilitation trajectories. Conventional protocols often rely on standardized movement patterns that may not fully engage target musculature, limiting therapeutic effectiveness. To address this, we developed a method that quantitatively analyzes motion dynamics and muscle activation to produce personalized movement trajectories. Multimodal data consisting of EMG signals from upper limb muscles, along with synchronized force and position measurements, were used to train a neural network capable of predicting activation levels from kinematic and dynamic inputs. This predictive model served as a surrogate for physiological response in a custom RL environment, enabling the PPO agent to discover trajectories that maximize activation of targeted muscles while maintaining biomechanical efficiency. Simulation results demonstrated that AI-optimized trajectories achieved significantly higher muscle activation compared to conventional paths. This integrated framework offers a scalable solution for adaptive rehabilitation planning, particularly when deployed within robotic systems. By enabling task-specific optimization grounded in patient-specific data, the approach advances the development of intelligent, individualized neurorehabilitation protocols that support more effective motor recovery.
| Original language | English |
|---|---|
| Title of host publication | ASME 2025 International Mechanical Engineering Congress and Exposition, IMECE 2025 |
| Volume | 4-A |
| DOIs | |
| State | Published - 2025 |
Fingerprint
Dive into the research topics of 'AI-DRIVEN TASK OPTIMIZATION FOR UPPER LIMB ROBOT-ASSISTED REHABILITATION'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver