One 'physics brain' can now inhabit and control any four-legged robot instantly, regardless of its limb length.
This world model disentangles environment physics from the robot's physical specs. It allows zero-shot control of new morphologies without the traditional weeks of retraining required when hardware changes.
Toward Hardware-Agnostic Quadrupedal World Models via Morphology Conditioning
arXiv · 2604.08780
World models promise a paradigm shift in robotics, where an agent learns the underlying physics of its environment once to enable efficient planning and behavior learning. However, current world models are often hardware-locked specialists: a model trained on a Boston Dynamics Spot robot fails catastrophically on a Unitree Go1 due to the mismatch in kinematic and dynamic properties, as the model overfits to specific embodiment constraints rather than capturing the universal locomotion dynamics.