Boston Dynamics’ Atlas humanoid robot perfects cartwheel and backflip

Boston Dynamics, in collaboration with the Robotics & AI Institute (RAI), has pushed its Atlas humanoid robot to perform advanced acrobatic maneuvers such as cartwheels and backflips, showcasing remarkable whole-body control and mobility. These feats were achieved through a simulation-trained, zero-shot learning framework that allows control policies developed in virtual environments to be directly applied to the physical robot without additional tuning. The robot demonstrated fluid, human-like walking, precise coordination during dynamic aerial moves, and the ability to recover from missteps, highlighting the sophistication and robustness of its control system. The RAI Institute, led by Boston Dynamics founder Marc Raibert, played a key role in developing the “Physical AI” that enables these capabilities. Their approach integrates all behaviors—ranging from walking to gymnastic maneuvers—within a single learning framework, aiming for generalist humanoid behavior capable of handling diverse tasks. This partnership, formalized in early 2025, focuses on advancing reinforcement