Bagel Labs is a physical AI research lab. Focused on building a model for autonomous robot control across tasks, environments, and robot types, designed to improve recursively through experience.
PARIS is Bagel’s training architecture. It lets different parts of a robotics model learn independently, then brings them together as one. In published tests with data and compute held equal, PARIS 1.0 improved results by 24%, and PARIS 2.0 by 50%.
WorldDiT is Bagel’s compact model architecture for robot learning and control. It sits on the reported Pareto frontier for model size and task success. It is the foundation for one model designed to improve across tasks, environments, and robot types.
