Personalized ML-based wearable robot control improves impaired arm function
James Arnold1, Prabhat Pathak1, Yichu Jin1
1John A. Paulson School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USA.
Nature Communications
|August 1, 2025
Summary
This study introduces a personalized control system for wearable robots that accurately detects user intention and improves upper limb movement in individuals with disabilities. The novel approach enhances arm function, mobility, and reduces compensatory movements.
Area of Science:
- Robotics
- Biomedical Engineering
- Rehabilitation Technology
Background:
- Wearable robots offer potential for upper limb assistance but face challenges in personalized control due to movement variability and support-transparency trade-offs.
- Effective control is crucial for real-world application of assistive robots for individuals with disabilities.
Purpose of the Study:
- To develop and evaluate a novel real-time controller for soft shoulder robots that integrates personalized machine learning intention detection and physics-based hysteresis modeling.
- To enhance upper limb function and movement quality in individuals with post-stroke and Amyotrophic Lateral Sclerosis (ALS) using an adaptive robotic assistance system.
Main Methods:
- Developed a personalized machine learning (ML) intention detection model using IMU and compression sensors to decode user motion intention.
- Implemented a physics-based hysteresis model for enhanced control transparency, adapted for practical real-world use.
- Integrated ML intention detection and hysteresis models into a real-time controller to modulate assistance based on user intention and kinematic state.
- Conducted a multi-day evaluation with 9 participants (5 post-stroke, 4 with ALS) using a soft shoulder robot.
Main Results:
- The controller achieved 94.2% accuracy in identifying shoulder movement from minimal angle changes.
- Arm-lowering force was reduced by 31.9% compared to a baseline controller.
- Significant improvements were observed in movement quality, including increased shoulder (17.5°), elbow (10.6°), and wrist (7.6°) range of motion (ROM).
- Trunk compensation was reduced by up to 25.4%, and hand-path efficiency improved by up to 53.8%.
Conclusions:
- The integrated control strategy effectively enhances upper limb function in individuals with post-stroke and ALS.
- Personalized intention detection and adaptive control are key to improving the performance and usability of wearable assistive robots.
- This approach demonstrates a promising advancement in robotic assistance for individuals with upper limb disabilities, leading to more natural and efficient movement.
