Interactive machine learning for personalised physiotherapeutic training

Body movement is our primary means of interacting with the world, and in future sensor-based modes of interaction, this will include AI-based functions.

This makes research into the design of movement-based AI critical, as technical systems literally shape users.The issue comes at an edge in physiotherapy.

Technology as a support for physiotherapy has been investigated for its potential of providing guidance and feedback in exercises to do at home.

While such exercises typically have normative stances on what is considered correct , in practice both what is considered correct, and how corrections are made, is situationally dependent. Current solutions often suffer from recognising only a very limited set of movements, or limit who can use the systems.

In this project, we investigate how machine learning can support the personalisation of out-of-clinic physical training.

We combine current research into open-ended tools for instructed physical training, with approaches to interactive machine learning to make tools adaptable to the individual.

Our goal is to empower both physiotherapists and patients to take control over their tools, and adapt them to suit their specific needs.

Of central concern is 1) maintaining the fluent concept of what is considered a ‘correct’ exercise execution and 2) to develop exploration mechanisms to control the ML-based learned representations, to ensure that physioterapists and patients can tailor them to their practice.