Figure's robots clean a room and make a bed in under 2 minutes

Tucking in a duvet is already annoying when you are alone. Doing it with two full-sized humanoids used to mean setting up heavy local networks, central servers, and millisecond-level position pings just to stop them from ripping the sheets in half.

Figure just bypassed that entire playbook.

In a new two-minute autonomous demo, two Figure F.03 humanoids walk into a bedroom, hang up clothes, clear surfaces, and make a hotel-neat bed without a human touching a controller. Figure CEO Brett Adcock called the finish better than what most humans manage.

No Wi-Fi chatter, just reading the room

The real breakthrough here is under the hood. Figure says this is the first demonstration of a single learned neural network powering multi-humanoid collaborative locomanipulation end-to-end, straight from camera pixels to physical actions.

Powered by Figure's Helix-02, the two F.03 units share a learned policy. They do not use a centralized orchestrator, and there is no explicit communication or wireless chatter passing coordinates back and forth. Instead, each robot coordinates purely through onboard visual cues, effectively reading the physical state of the room and the other machine in real time.

Figure also showed off the F.03 hardware package. When the shift wraps up, the humanoids step onto a wireless foot-charging pad that pumps power at 2 kW, eliminating the need for human plugging or awkward charging docks.

Why it matters

We think this demo highlights an important architectural bet on multi-agent robotics. Most multi-robot setups in factories rely on rigid central planners, which crumble the moment environmental variables shift or Wi-Fi drops. Figure is betting on emergent coordination: two bodies sharing one visual instinct.

If this scales beyond scripted bedroom stages, it completely rewrites the economics of home and hospitality robotics. Cleaning a hotel room or an eldercare facility is rarely a solo job. Showing that two robots can tackle soft, deformable objects together in under two minutes without a brittle network stack makes fleet deployment look far more practical.

The catch, of course, is reliability in the wild. Bedding comes in countless weights, room layouts are messy, and relying purely on visual cues means a single lighting glitch or occlusion could cause both arms to pull in opposite directions. Figure proved the brain works in the lab; the real test is whether these robots can handle a messy Airbnb without tearing the blankets.