AI MODELS FOR MOVEMENT

We are building AI models for movement to fundamentally advance our ability to understand human movement and, more importantly, to reduce movement impairments. 

Impact

The potential impact of AI models for movement for rehabilitation is immense.

Rehabilitation could be more personalized, accessible, and effective with data-driven tools to help prevent, detect, diagnose, and treat movement impairments, and assess rehabilitation outcomes. Low-cost and accessible tools to measure movement would enable the large-scale studies necessary to uncover the multifactorial contributors to limited mobility. 

Our models and methods are guided by the needs of several driving rehabilitation challenges. We are developing:

• A tool to optimize an individual’s personal movement pattern to help decrease pain and improve function in those with knee osteoarthritis
• An accurate, interactive tool to guide surgical planning for children with cerebral palsy
• A rapid, low-cost tool to identify modifiable risk factors for anterior cruciate ligament (ACL) injury or reinjury

Some of these models will be foundation models.

What is a foundation model?

Historically, machine learning models have been designed to estimate a specific output (e.g., a freezing of gait event) given a set of inputs (e.g., a trajectory of wearable sensor data). In contrast, new generative models are trained to generate instances of the data (e.g., a passage of text), sometimes based on an input prompt or set of related data. The large language models behind platforms, like ChatGPT and Claude, are generative models.  

Generative models have proven to be incredibly powerful, achieving high levels of performance with flexible inputs and outputs and on tasks the models were not specifically trained to perform. Given these broad capabilities, this new class of models has been termed “Foundation Models.”

Rehabilitation research has not had its “ChatGPT moment” yet, and there exists no foundation model for human movement. The FAIR Center is filling that gap.