Breakthroughs in lower limb exoskeletons that use deep learning have led to exoskeletons being able to provide personalized assistance, particularly for senior individuals. On 5 March 2025, the AI for Good platform convened a webinar showcasing research from Dr. Aaron Young, Associate Professor and Woodruff Faculty Fellow at Georgia Tech’s Woodruff School of Mechanical Engineering, and Director of the Exoskeleton and Prosthetic Intelligent Controls Lab. The webinar spotlighted the ability of temporal convolutional neural networks that can process real-time sensor data such as encoders, joint angles, velocities and inertial measurement units at 200 Hz to directly predict internal joint movements. This new breakthrough can support dynamic torque adjustments in exoskeletons, without requiring task-specific calibration. Promisingly, this technique resulted in a 14 per cent reduction in metabolic cost during walking and lifting tasks, exceeding standard State-based controllers by adapting to each user’s unique biomechanical characteristics.
The utilization of open-source datasets enables task-agnostic control for 29 different activities, including walking, ramp ascent and descent, and stair climbing. The webinar noted that this eliminates the need for manual controller calibration, while staying in sync with user intent by concentrating on internal physiological cues. While the webinar acknowledged current challenges such as device weight and the complexities of real-world deployment, it also pointed to promising future directions for lighter, clothing-integrated systems with broader clinical applications, including post-stroke rehabilitation. These advancements make a significant contribution to wearable robots by addressing the challenge of generalizing across numerous activities while overcoming the limits of standard control systems. They also open the way for more user-friendly, plug-and-play exoskeletons that can be effortlessly integrated into daily life, allowing individuals to move freely and independently.
This case study is an excerpt from the AI for Good flagship report produced by UNU-CPR, Unlocking AI's Potential to Serve Humanity.
