Helix, humanoid robotics specialist Figure AI’s Vision Language Action (VLA) model, has shown it can “adapt to dramatically different real-world challenges with nothing more than new data”.
After folding laundry and rearranging packages, Helix is now taking on another everyday task: loading a dishwasher.
At first glance, loading a dishwasher sounds like a simple task – just pick up each object and place it in the dishwasher.
But in reality, dishwasher loading bundles together many difficult problems in robotics: dishes often need to be isolated from cluttered stacks, reoriented, or handed off between two arms working in sync; slippery or fragile items demand fingertip-level precision; and dishwasher racks provide only centimeter-scale tolerance for error.
On top of that, every load is different – novel objects, messy starting states, and unexpected collisions mean the system must constantly adapt and recover while maintaining robust performance.
Key results
The same Helix model that folded towels and sorted packages can now load a dishwasher. No new algorithms, no special-case engineering, just new data.
With that, Helix learned to:
- Singulate stacked plates and load them in order.
- Pick up a glass with one hand, reorient it, and place it carefully with the other.
- Adjust its strategy to account for messy starting configurations.
- Recover gracefully from errors like misgrasps or collisions.
Why this matters
Dishwasher loading, package logistics, and towel folding may seem worlds apart, yet Helix handles all with the same general-purpose architecture.
This represents another step toward scalable humanoid intelligence: a single system that can incrementally learn new capabilities with additional data, building toward broader applicability without special-case engineering.
