FDT-BioTech: Biosignals Fusion-Informed Neuromusculoskeletal Reconstruction in Immersive Digital Twin for Personalized Stroke Gait Rehabilitation Using Lower-Limb Exoskeletons

Stroke can weaken one side of the body and make walking slow, uneven, and unsafe. Wearable robotic exoskeletons may help, but their support must be tuned or adjusted for each person. This project will create a digital twin, a computer model that stays linked to a person and a wearable lower-limb exoskeleton. The model will use muscle signals, such as ultrasound images and muscle activations, body motion, foot pressure, and robot sensor data. It will estimate how the person is moving and how the muscles are working. It will also test possible robot settings in a virtual space before those settings are used on the person. This approach may make exoskeleton training safer, quicker to adjust, and better matched to each stroke survivor. It may also help researchers test new rehabilitation devices with fewer risky or costly physical trials. College students will join the work through research training, class projects, senior design, and summer research. Middle and high school students will learn about rehabilitation robotics through lab tours and hands-on demonstrations. The project will develop a digital twin that accounts for uncertainty while guiding control of a wearable lower-limb exoskeleton. First, physics-based and data-based models will combine electromyography, which measures muscle electrical activity, with sonomyography, which uses ultrasound images to track muscle shape and motion. These models will build a person-specific view of muscle and movement states. They will also create synthetic data for model training and testing. Second, the digital twin will represent the person, the exoskeleton, and the walking setting. It will detect changes in gait, device alignment, task demands, and possible hazards, such as low toe clearance or unsafe foot placement. Virtual “what-if” trials will screen robot assistance plans before they reach the physical device. Third, a hybrid online-offline reinforcement learning method will tune a small set of control settings, such as assistance level and timing. Fast safety controls will stay on the exoskeleton. Slower digital-twin updates will improve tuning between walking trials. Verification, validation, and uncertainty quantification will test model accuracy, measure confidence, and define when the system should update, pause, or return to a safe setting. The team will evaluate the framework through staged computer, hardware, and human tests, including tests with adults who have chronic stroke. Outcomes will include gait endurance, symmetry, stability, muscle response, comfort, and safety. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.