Computer vision-informed machine learning for user-specific exoskeleton control

Team Members

Changseob Song

Bogdan Ivanyuk

Adrian Krieger

Kolin Huang

Sunny Kim

Runqiu Ye

Joonyoung Huh

Annika Lee

Diya Dinesh

The field of exoskeleton control has long faced challenges in achieving rapid personalization and adapting systems to diverse tasks without requiring large amounts of data. Traditional methods often rely on extensive clinical data collection and manual adjustments, which can be time-consuming and inefficient. Addressing these gaps, our team will focus on the rapid personalization of exoskeleton control optimization, leveraging minimal video data and physics-informed simulations. Our approaches will emphasizes cross-task adaptation, allowing the system to effectively adjust to a diverse range of task sets through the generation of synthetic data. Additionally, the method prioritizes patient-specific assistance, optimizing support based on individual needs while minimizing the amount of clinical data required for the process. Our strategy combines efficiency and personalization, streamlining the development of exoskeleton systems for various applications, namely translation out of the lab environment and into daily ambulation environments. 

Related Work

Personalization of Joint Kinematic Estimation Using Computer Vision