AI Model Learns to Predict Human Gait for Smarter, Pre-Trained Exoskeleton Control

Thanks to new AI training methods, wearable exoskeletons—robotic suits that help people with mobility impairments—are now making significant strides. In the past, creating each design required laborious laboratory work, gathering user motion data to train the controller of the suit. Scientists at Georgia Tech now have an AI technique that pre-trains exoskeleton controllers using current human motion data. Long-term retraining is eliminated, and real-world deployment becomes more feasible.

AI-driven training cuts development time

According to the published study, researchers adapted a deep learning model (once used to transform images) to translate normal walking motions into what they would look like if the person wore an exoskeleton. It predicts hip and knee sensor readings and learns how much robotic assistance is needed. The result is a pre-trained exoskeleton “brain” that matches the performance of lab-tuned controllers. In tests on a leg exoskeleton, the AI controller tracked the user's joint movements and boosted their effort by up to about 20%.

Real-World Benefits and Future Prospects

The new approach allows designers to produce and test prototype exoskeletons much more quickly by avoiding repeated lab training. A startup could go through several device iterations without gathering new data, according to Prof. Young. This type of controller may also be used to power prosthetic limbs or upper-body exoskeletons. Exoskeletons could be used to help amputees or stroke survivors regain their mobility, as well as to provide factory workers with additional strength and lessen fatigue during lifting tasks. All things considered, AI-powered training might eventually make these science-fiction-like gadgets useful.