Optimizing Exoskeleton Control Parameter

Physiological state-based exoskeleton controllers, such as those that directly estimate a user’s biological joint moments from onboard wearable sensors, hold promise for enabling task-agnostic and generalizable control. In our approach, we leverage deep neural networks to handle this estimation, allowing the exoskeleton to respond fluently and naturally to the user’s intended motion, whether in cyclic or non-cyclic tasks.

However, the critical step of mapping the estimated state (i.e., joint moment) to effective exoskeleton torque remains underexplored. Most current approaches rely on a static, single scaling value, which is suboptimal for capturing the dynamic and context-dependent nature of human locomotion. Fully resolving this mapping challenge is key to achieving robust performance across tasks and environments.

Overall, our goal is to develop task-generalizable control strategies that are seamless, adaptive, and effective. We view physiological state tracking as a powerful starting point. By integrating deep learning, wearable sensing, and biomechanical modeling, we aim to advance adaptive control frameworks that significantly improve exoskeleton-assisted locomotion.

Team Members

• Jimin An