Move over laundry—Figure’s humanoid robot has just leveled up again. It can now load a dishwasher.
Well… assuming you already own one.
The robot still runs on the Helix architecture, the same system that mastered towel folding and package sorting. No new algorithms, no special engineering—just new data.
Online chatter quickly spotted an unexpected concern:
“Wait… does it wash its hands after touching dirty plates? Would it use the same hands to fold clothes?”
Apparently, hand hygiene is a bigger topic than we thought.
From Package Sorting to Household Chores
Three months ago, the Figure robot wowed the world by sorting packages for an entire hour, recognizing, grasping, flipping, and transporting items with near-human speed and flexibility.
Last month, it learned how to fold towels and clothes. Now, the same Helix system is taking on a completely different task: loading a dishwasher.
At first glance, this may seem simple: pick up dishes, place them inside. But the reality is far more complex:
- Dishes must be separated from messy stacks and reoriented.
- Two-arm coordination is often required.
- Smooth or fragile items need fingertip-level precision.
- Dishwasher racks tolerate only centimeter-level errors.
Every loading scenario is different: new dishware, disorganized placements, unexpected collisions. The system must adapt, correct, and maintain stability in real-time.
Like its previous achievements, the Figure 02 accomplishes this without new algorithms or engineering tweaks—it learns purely from data.
Helix Skills in Action
Through dishwasher loading, the Helix architecture has learned to:
- Separate stacked plates and place them sequentially.
- Pick up glasses with one hand, rotate them, and carefully set them with the other.
- Adjust movements based on chaotic starting positions.
- Recover smoothly from failed grasps or collisions.
Even this “simple” task expands the system’s capabilities considerably.
The Helix Architecture: A Universal Humanoid Brain
Helix is Figure’s first major system post-OpenAI, an end-to-end Vision-Language-Action (VLA) model that allows humanoid robots to perceive, reason, and act like humans.
It consists of two interconnected systems trained end-to-end.
A single unified model can perform a wide variety of tasks:
- Pick and place objects in different containers.
- Operate drawers and refrigerators.
- Coordinate multi-robot handoffs.
- Handle thousands of novel objects with precision.
Dishwasher loading, package sorting, and towel folding may seem unrelated—but Helix handles them all with the same architecture. Simply adding new data teaches it new skills without engineering changes.
This marks a major milestone in scalable humanoid intelligence, bringing Figure robots closer to performing a full range of household and industrial tasks.
References:
[1] https://x.com/Figure_robot/status/1963266237426979300
[2] https://www.figure.ai/news/helix-loads-the-dishwasher