Figure AI has unveiled Helix 02, a leap forward for humanoid robotics that expands intelligence from partial motion control to seamless command of the entire body. The platform merges locomotion, balance, and manipulation into one neural system, allowing robots to function smoothly across complex environments.
Rather than separating perception, planning, and movement, Helix 02 operates through an end-to-end pipeline. Camera input is translated directly into motor action, giving the robot the ability to perform extended activities continuously, without resets or operator input.
In one demonstration, the humanoid navigates a kitchen while unloading and reloading a dishwasher for several minutes. The task combines walking, reaching, grasping, and stabilizing—showcasing a degree of persistence and coordination that pushes the boundary of autonomous performance.
The system’s architecture is built on total sensor fusion. Vision, touch, and proprioceptive awareness are processed together inside a single visuomotor brain that controls every joint.
Driving this capability is “System 0,” a learned controller derived from extensive human motion capture and simulation-to-real reinforcement learning. By embedding movement knowledge into neural priors, the approach dramatically reduces the need for hand-coded instructions.
Enhanced hardware on Figure 03, including tactile sensing and palm cameras, unlocks delicate operations such as manipulating small items or extracting objects from clutter.
Together, these advances mark a step toward robots that behave with increasing fluidity, adaptability, and human-like coordination.