Collaborative Clean: Figure Demonstrates Multi-Robot Bedroom Reset via Helix 02

Key Takeaways

Show all

• Two Figure 03 robots successfully reset a bedroom in under two minutes, performing tasks ranging from opening doors to collaborative bed-making.
• The demonstration utilizes a single learned Helix 02 neural network with no central coordinator or message passing between the robots.
• New capabilities include complex locomanipulation tasks like operating a foot-pedal trash can while balancing on one leg.
• This milestone advances Figure’s goal of domestic utility, following its recent $400–$600 monthly lease pricing strategy for home assistants.

Following a brief social media tease, Figure has unveiled its most sophisticated domestic demonstration to date: two Figure 03 (F.03) humanoids autonomously tidying a bedroom in a coordinated effort. The "Helix-02 Bedroom Tidy" marks a significant leap from previous living room demonstrations, moving beyond single-robot tasks to complex, multi-agent collaboration.

The Decentralized Brain

The most striking technical detail of the demonstration is the lack of a central "overseer" or shared planner. While multi-robot systems often rely on a central server to assign tasks, Figure claims these units operate as independent agents. Each robot utilizes its own cameras to infer its partner's intent through motion alone—mirroring how two humans might instinctively coordinate while folding a sheet.

This is powered by the Helix 02 architecture, a Vision-Language-Action (VLA) "omni-model" that Figure has been refining to replace over 100,000 lines of hand-coded C++. By computing torque directly from pixels, the robots can react in real-time to the "shared goal" without explicit communication.

A "Boss Battle" for Locomanipulation

The two-minute "bedroom reset" sequence requires the robots to navigate what CEO Brett Adcock calls the "chaos and entropy" of a home environment. The demonstration highlights several new learned behaviors:

• Single-Leg Balance: One robot is shown picking up trash and shifting its weight to one leg to depress a trash can’s foot pedal with the other—a feat of dynamic balance that builds on Figure’s Never Fall protocol.
• Complex Tool Use: The robots open doors with whole-body coordination, hang clothes on narrow fixtures, and reorient headphones mid-air to place them on a stand.
• Deformable Object Handling: Making the bed is described as a particularly difficult challenge because the comforter has no fixed geometry. The robots must constantly update their predictions as the fabric folds and slides under shared tension.

Closing the "Janitor Gap"

As Figure continues to scale production at its BotQ facility to one robot per hour, the focus has shifted toward generating the "interaction data" needed for general-purpose utility. Adcock has previously suggested that a car-like $400 to $600 monthly lease could bring these machines into homes for 24/7 assistance.

However, despite the "human parity" claimed in industrial package sorting, the home remains the ultimate test. Adcock remains cautious, noting that he still "babysits" the machines when they are around his own children while the company works to solve the "long tail" of edge-case failures.