Figure 02 Becomes First Humanoid to Autonomously Fold Laundry

• Figure’s humanoid robot, powered by Helix, folded towels using only a new dataset with no changes to its AI model.
• The robot handled tasks like tracing edges, pinching corners, and untangling towels while keeping natural gaze and gestures.

Figure’s humanoid robot, Figure 02, has shifted from logistics to folding laundry, one of the most demanding dexterous manipulation tasks, and a first for a humanoid robot. For an hour straight, it folded towels on its own. Towels are soft and unpredictable, forcing the robot to constantly adjust and use precise finger control to grip edges, smooth them flat, and keep the fabric from bunching or slipping.

In its latest demonstration, powered by the Helix Vision Language Action (VLA) model, Figure 02 autonomously picked towels from a mixed pile, adjusted folding strategies for varying starting configurations, recovered from multi-pick errors, and unraveled tangled items before completing folds. The same Helix system that handled package reorientation in logistics was used for laundry folding with no changes to its design or settings.. Only the dataset was new.

Helix doesn’t rely on detailed object models, which often fail with soft, changing items like towels. It goes straight from vision and language to precise movement. This shows that the same physical platform and general-purpose architecture can move from logistics to household tasks and will get faster and more capable as it trains on more real-world data.

🌀 Tom’s Take:

The big leap here is that Figure 02 learned laundry folding without changing its AI design, just by feeding it new data. That means the same robot brain can pick up completely different skills fast.

Source: Figure