University of Michigan
The goal of this clinical trial is to test a new, impairment-aware robotic control software framework to see if its smart adaptation can improve walking recovery in healthy adults and chronic stroke survivors. . The main questions it aims to answer are: Can the new control software safely use sensors and machine learning to predict and instantly adapt to a user's specific walking needs? Does training with a robotic device driven by this new adaptive control framework improve walking speed and overall mobility in stroke survivors? Researchers will compare a lower-limb orthosis operating under the new "smart" control software (which adapts to the user's impairment) to the same device operating under a standard, non-adaptive controller (which uses rigid or fixed assistance) to see if the new control approach leads to greater improvements in walking ability. Participants will: Walk on treadmills, flat walkways, or stairs while wearing a robotic leg orthosis driven by the different control software systems being tested. Wear small tracking tools (like reflective motion-capture markers and muscle activity sensors) so researchers can precisely measure how their movements interact with each control program. Complete standard walking tests to measure their walking speed and overall mobility under each software condition.
Inclusion Criteria: * Cohort 1: Able-Bodied Participants (Initial Validation) * Healthy young adults. * No history of neurological, orthopedic, or cardiovascular impairments affecting gait or balance. * Able to walk independently without assistive devices. Cohort 2: Stroke Survivors (Clinical Efficacy Pilot) * Individuals with a documented history of chronic stroke. * Persistent unilateral lower-limb motor impairment resulting in a pathological gait pattern (heterogeneous gait deficits). * Stable medical condition allowing for participation in intensive physical rehabilitation tasks. * Able…
An AI-driven, machine learning-based control software integrated into a wearable lower-limb powered orthosis. The system utilizes a Bayesian Neural Network (BNN) to analyze a user's pathological walking patterns (kinematics) in real-time via onboard sensors. Based on this real-time performance, the device dynamically modulates its physical assistance along a seamless continuum. It automatically transitions between stiff corrective guidance (position-based gait training) when the user struggles, and compliant, volitional torque support (torque-based assistance) as the user's independent walking ability improves.
A standard control paradigm for lower-limb powered orthoses that provides non-adaptive physical assistance during gait training. Depending on the trial block, the device operates in one of two static modalities: either rigid position-based gait training (GT) that physically guides the patient's limbs through a fixed, predetermined trajectory regardless of effort, or torque-based volitional augmentation (VA) that proportionally amplifies existing muscle output or ground reaction forces. Unlike the experimental intervention, this controller cannot interpret kinematics in real-time or dynamically modulate assistance along a continuous spectrum based on the user's instantaneous performance.