Figure Helix: a Vision-Language-Action model for a humanoid's whole upper body
In one sentence On February 20, 2025 Figure unveils Helix, a generalist VLA model that controls the entire upper body of the Figure 02 robot at 200 Hz (35 degrees of freedom, fingers included), runs fully on onboard GPUs and coordinates two robots on the same task. Demo: putting away groceries with never-seen items.
Two weeks after ending its collaboration with OpenAI (February 4, 2025), Figure unveils Helix, its Vision-Language-Action (VLA) model: a network that takes images and natural-language instructions and directly outputs the robot's movements.
Helix controls the whole upper body of the Figure 02 (wrists, torso, head and individual fingers, 35 degrees of freedom) at 200 Hz. Two robots can run the same weights at once and collaborate on a task, for example handing over a bag of cookies.
In the demos the robots put away groceries with items never seen in training and pick up thousands of different household objects in clutter, from verbal instructions alone.
Companies
Figure
Tools
Helix, Figure 02
Tags
Sources