PRISM is a robotics research project for learning Polynomial Representations for Interaction-Structured Motor Control. It makes action-relevant interactions among observable robot signals explicit, compact, and learnable, improving control without adding new sensors.
The method adds polynomial interaction features to policy learning so controllers can capture products and relationships among state variables. Experiments cover humanoid locomotion, low-friction tracking, and contact-rich manipulation, showing improvements over standard and larger control baselines.
PRISM is useful for robot learning researchers, motor-control engineers, and teams building policies for complex physical interaction. It offers a structured representation layer that can reveal physical relationships while improving policy performance in challenging dynamics.
Get more likes & reach the top of search results by adding this button on your site!
Get top updates in AI to your inbox every weekend. It's free!