UC San Diego's task-agnostic exosuit controller cuts torque by 73%

Researchers at UC San Diego's Soft Robotics and Bioinspired Robotics labs have unveiled a control strategy that overcomes a longstanding bottleneck in soft exosuits: delivering responsive, intention-driven assistance without bulky physiological or force sensors. The team — Anoush Sepehri, Zachary Huang, Raymond de Callafon, Michael T. Tolley, and Tania K. Morimoto — built an inverse-plant controller for pneumatically actuated soft exosuits that uses only kinematic data (joint angles) to infer user intent. At the core is a Hammerstein dynamic model that combines a Preisach hysteresis model (capturing the actuator's nonlinear static behavior) with a linear time-invariant filter (representing its dynamic response). This allows the system to predict how the suit will react and then invert that model in the control loop for real-time, task-agnostic assistance. Crucially, the model is personalized with just 140 seconds of per-user calibration data, making it practical for real-world deployment.

In tests on a rig that emulated a soft wrist exosuit, the controller reduced interaction torque by up to 73% and lowered activation of key flexor and extensor muscles by as much as 47%, compared with no assistance, across joint velocities from 8°/s to 120°/s. These gains held without any pre-defined task templates or force/EMG sensors, meaning the suit can smoothly assist across the wide range of unscripted motions typical of daily living. The work positions soft exosuits as a viable option for rehabilitation, augmentation, and assistive daily-living devices — not just in controlled clinical settings but in the messy real world. By decoupling assistance from task-specific programming, this control strategy could accelerate adoption of wearable robots in home care, stroke rehab, and occupational support. The full paper is available on arXiv.