Mocobots Learn the Physics of Any Payload—On the Fly

A team of mobile cobots, or "mocobots," can now autonomously learn how to jointly handle unknown payloads—without prior models or external sensing. By first randomizing motion to infer grasp frames, then using force data to estimate mass properties, and finally leveraging dynamic motion to identify full inertia, the system builds a complete physical model of the object on the fly.

Validated on compliant Omnid cobots, the approach achieves high-accuracy estimates of grasp geometry, mass distribution, and inertia, enabling more precise and coordinated manipulation of bulky, unfamiliar objects in real-world collaborative robotics scenarios.