Boston Dynamics gives Atlas robot roller-skates for mind-bending control tests

Boston Dynamics engineers performed a fascinating experiment by attaching wheels to the feet of their Atlas humanoid robot. They combined traditional physics-based control models with reinforcement learning to allow the robot to navigate unstructured terrain on wheels, treating it as a research testbed rather than a commercial product.

The goal was to push Atlas's actuation limits and explore how to handle unpredictable real-world conditions that are impossible to perfectly model. By processing real-time feedback, the robot could make dynamic adjustments to its pose and force, bouncing back from disturbances. This represents cutting-edge progress in marrying deterministic control with adaptive AI for robust, whole-body robotics.