Adaptive Reinforcement Learning for Soft Exosuit

Soft exosuits have the potential to improve mobility and reduce physical strain during walking, but providing effective assistance across different users and walking conditions remains challenging. Traditional learning control systems often require extensive tuning, large amounts of training data, or simulation environments that may not accurately reflect real-world movement. Variations in user gait, terrain, and walking intensity can also reduce the effectiveness of generalized assistance strategies. This creates a need for adaptive and personalized control methods that can safely optimize assistance in real time.

 

Researchers at Arizona State University have developed an innovative Online Adaptation from an Offline Imitating Expert Policy (AIP) framework to control soft exosuits. The approach combines offline expert policy imitation with online actor-critic reinforcement learning to personalize robotic assistance during walking in real time. The system incorporates safety constraints and data quality improvements to ensure stable and reliable operation across diverse users and conditions, including incline walking. Experimental validation demonstrates reduced muscle effort and improved walking patterns without relying on simulation-based training. This enables efficient, real-time adaptation of exosuit assistance in practical settings.

This technology presents a novel reinforcement learning framework for personalized, safe, and efficient soft exosuit assistance with human walking.

 

 

For more information about this opportunity, please see