Close project

AidWear: AI-driven wearable robots

The AidWear project aims to develop the artificial intelligence frameworks that are necessary to enable Robotic Assistive Devices (active prosthetics and lower-limb exoskeletons) that give Parkinson’s patients and individuals with an amputation a better quality of life. Building on the results of the AI4exo project and taking advantage of existing hardware, AidWear will advance three areas of interest: intention detection, mid-level optimization, and dynamic simulation.

Accurate intention prediction is an important requirement for the (high-level) controllers of wearable robots, as it enables them to proactively respond to the user and provide appropriate assistance when needed. Current SotA intention prediction algorithms are, however, unable to handle the variability of everyday gait at home. To solve this problem, we will develop a multimodal framework that exploits kinematic data with measures of intent (i.e., egocentric video and eye tracking). Furthermore, SotA algorithms require substantial patient cohort-specific development data. We will therefore develop a framework for intention prediction driven by continual learning that can efficiently adapt to new patient cohorts, while reducing the amount of patient cohort-specific development data. Finally, we will develop an approach that automatically simplifies large intention prediction models so that they can run in real-time on resource-constrained computing latforms.

Another issue pertains to the mid-level control of these devices. Typically, the torque profiles for the motors are based on data from able-bodied subjects, who walk very differently than gait-impaired subjects. AidWear proposes human-in-the-loop optimization, i.e., the use of AI to optimize control parameters based on the measured response from the user, to achieve assistance profiles tailored to the gait impairment and even the specific user.

Finally, the machine learning-based approaches proposed in AidWear require large amounts of experimental data from the prospective user, which are very costly to obtain. Numerical modelling and simulations can play a pivotal role in solving this issue. However, existing software falls short in representing a human body with robotic assistive devices. Therefore, we will develop a novel framework for numerical modelling and simulation of the human musculoskeletal system that streamlines the design and optimization of robotic assistive devices.

The project will generate international exposure for Belgian AI and robotics through participation in the 2024 CYBATHLON competition. Furthermore, there are concrete paths to provide a return to society, such as technology transfer to existing Belgian start-ups, reduced healthcare costs for two large patient groups, and dissemination activities to showcase the potential of AI and robotics in healthcare.

This project is made possible by the Federal Public Service for Policy and Support.

© 2024 - Vrije Universiteit Brussel - Dept. MECH - All rights reserved