Human musculoskeletal models
Contact person: Anders Malthe-S?renssen
Keywords: machine learning, curriculum learning, musculosekeletal models, control theory
Research groups: Center for Computing in Science Education / TechWell Gemini Center
Department of Physics
Recent advances in hybrid physics-AI models and fundamental knowledge of human physiology allow the creation of detailed musculoskeletal models of humans. Such models are important for applications in medicine, healthy aging and product design, while raising fundamental questions in sensorimotor control and machine learning, often posing competitions for conferences such as NeuroIPS. These models have many internal degrees of freedom, but effective representations and original training methods simplify representation, training and applications. This theme focuses on combining advanced computational modeling and machine learning models to develop musculoskeletal models and to train their efficient control. Results can be tuned through close collaboration on experiments using a unique 3d camera technology for marker free tracking.
Topics from methodological research:
- Curriculum learning combined with reinforcement learning for musculoskeletal models
- Effective latent space representations for dimensionality reduction in sensorimotor control
- Continual learning methods for gradual development of representations through drills
Topics from natural sciences or technology:
- Musculoskeletal models
- Personalized physiological avatars
- Healthy aging
External partners:
- EPFL
- SINTEF
- Flokk (Norwegian design company)
Mentoring and internship will be offered by a relevant external partner.
