Figure AI has demonstrated a significant breakthrough in robot autonomy: two humanoids successfully completed a full bedroom reset, including making a bed together, using only visual cues to coordinate their actions without any explicit communication between them. The robots, powered by Helix-02, a vision-language-action system that processes raw camera input directly into motor commands, read each other's movements through head nods, stance shifts, and arm angles to stay synchronized. This marks what the company calls the first instance of a single learned neural network performing collaborative tasks across multiple humanoid robots.

How Do These Robots Understand Each Other Without Talking?

Unlike traditional robotic setups that rely on separate planners, message passing, or a central coordinator, Figure's F.03 humanoids operate on a fundamentally different principle. Each robot has its own camera and infers its partner's intent purely from motion. The Helix-02 system runs entirely on edge computing, meaning the inference happens locally on each robot's onboard computer without requiring network connections or external coordination signals.

The bedroom demonstration required the robots to handle several complex, interdependent tasks simultaneously. They opened doors, hung clothes on a coat tree, put away headphones, closed a book, took out rubbish, pushed an office chair under a desk, and then worked together to make the bed. The bed-making sequence was particularly demanding because it involved lifting, unfolding, spreading, folding, and smoothing a duvet while correcting wrinkles and bunched edges as the fabric settled.

What Makes Multi-Robot Coordination So Difficult?

• Interdependent Problem Solving: Two humanoids in one room are not simply two single-robot tasks running in parallel. Every move one machine makes changes the problem the other must solve, requiring constant real-time prediction and adaptation of the partner's next action.
• Deformable Object Handling: The duvet has no fixed shape, no rigid geometry, and no natural divide between "your half" and "mine." Each robot commits to a contact point while predicting what its partner will do, updating those predictions tens of times per second as the fabric folds, drapes, and slides under shared tension.
• Speed and Seamless Transitions: The entire sequence runs in under two minutes, requiring the robots to walk naturally between locations, balance dynamically on one leg to operate a pedal bin, and switch seamlessly between rigid, deformable, articulated, and collaborative manipulation without scripted handoffs between subtasks.

Brett Adcock, Figure AI's Chief Executive, emphasized the significance of this achievement.

"There is no explicit messaging between these robots; they coordinate their actions fully visually, e.g. head nods," Adcock stated.

Brett Adcock, Chief Executive at Figure AI

The task was fully autonomous and ran at normal speed with no teleoperation and no human intervention. This stands in contrast to earlier skepticism about Figure's livestream demonstrations, where critics questioned whether head gestures indicated hidden human control. The company has since clarified that such movements are natural byproducts of the Helix-02 whole-body controller clearing arm pathways during normal operation.

How Does This Connect to Figure's Broader Roadmap?

The bedroom coordination demonstration feeds directly into Figure's larger vision for generalized autonomy. The underlying Helix-02 system was not built specifically for bedrooms; it is a single learned policy that expands its skills as it is fed more data. Earlier this year, the same approach allowed a Figure robot to load a dishwasher in a full-sized kitchen in four minutes, and in March, a solo F.03 tidied a living room, spraying and wiping surfaces, sorting toys, and replacing cushions on a sofa.

Figure is now leveraging a newly deployed cluster of Nvidia Blackwell B200 graphics processing units (GPUs) to train its largest AI models, targeting true generalized autonomy in unseen environments. The company also launched a dedicated 70-person AI laboratory called HARK last summer, which has already deployed an onboard, real-time speech-to-speech voice model across the current fleet.

The company's next-generation platform, Figure 4, represents a ground-up architectural redesign optimized for data collection and featuring a watchmaker-grade humanoid hand with more actuators than the entire rest of the robot's body combined. Adcock described the upcoming machine as "unrecognizable" from prior iterations, comparing its impending launch to the industry's "iPhone 1 moment".

Figure is also aggressively onshoring its production to eliminate geopolitical risks. The company forecasts zero supply chain exposure to China by next quarter, having successfully onshored or diversified the production of its custom motors, gearboxes, sensors, and printed circuit boards. Operating from its BotQ manufacturing facility on the Figure campus, the company is on track to manufacture between 60 and 70 humanoid robots per week, representing an annual production run rate in the thousands.

The company describes the multi-robot demonstration as an important first step toward a future in which intelligent humanoids routinely work together in homes, warehouses, and factories, handling shared goals in spaces where people, objects, and other machines are constantly on the move.