Figure humanoids make bed, tidy room
- Figure showed two Figure 03 humanoids resetting a messy bedroom on May 8, with Helix 02 handling the whole task autonomously. - The standout detail is the bed-making: the robots coordinated visually, not through a shared planner, and finished the full room reset in under 2 minutes. - It matters because home cleanup is messy, soft-object work — exactly the kind of unscripted manipulation humanoids usually struggle to do.
Humanoid robots are usually shown doing one neat trick at a time — pick up a box, carry a tote, maybe sort parts in a factory. A bedroom is meaner. Clothes flop around, blankets deform, furniture blocks paths, and two bodies can get in each other’s way fast. That is why Figure’s new demo lands: on May 8, the company showed two Figure 03 robots tidying a bedroom together, including hanging clothes and making a bed, with no direct human control. ### What actually happened? Figure released a “Helix-02 Bedroom Tidy” demo showing two humanoids reset a staged messy room in under 2 minutes. The robots picked up objects, opened and closed doors, hung clothing, moved items into place, and jointly spread and smoothed a comforter over a bed. ### Why is bed-making the hard part? A bed is a soft-body manipulation problem. The blanket does not keep a fixed shape, the edges slip, and each pull changes the next grasp. (figure.ai) Now add a second robot. Each machine has to infer where the other is going, avoid fighting the fabric, and keep tension balanced — more like two people handling a large sheet than a robot placing a rigid object. ### How are the robots coordinating? The interesting bit is that Figure says they are not using a central overseer. The robots coordinate through visual observation — basically watching the room, the objects, and each other, then updating actions in real time. Figure describes Helix 02 as a full-body vision-language-action system that runs from onboard sensing and controls walking, balance, and manipulation as one loop. (businessinsider.com) ### What changed from earlier demos? This is not Figure’s first home-cleanup video, but it is a step up. In January, Figure introduced Helix 02 as a system that could control the whole robot from pixels. In March, the company showed one robot tidying a living room. The new bedroom demo adds two-robot coordination and more soft-object handling, which is a tougher version of the same general problem. (letsdatascience.com) ### Why does “from pixels” matter? Because the old way in robotics is a stack of separate modules — one for walking, one for grasping, one for planning, then a bunch of brittle handoffs. Figure’s pitch is that Helix 02 skips much of that stitching and learns a unified control policy from sensor input to motor output. The company says System 0, its learned whole-body controller, replaced 109,504 lines of hand-engineered C++ and was trained with more than 1,000 hours of human motion data plus sim-to-real reinforcement learning. (figure.ai) ### Is this a real product moment? Not yet. It is still a curated demo from the company that built it. But it is a better demo than the usual robotics sizzle reel because the task has clutter, deformable objects, long action chains, and another moving agent in the loop. Those are exactly the conditions that break scripted systems. ### So what should you take from it? The point is not that your next apartment comes with two robot maids. (figure.ai) The point is that humanoid control is inching from isolated tricks toward messy household sequences. If Figure can keep adding tasks without rebuilding the software stack each time, that is the real unlock — not one made bed, but a general system that can survive the chaos of ordinary rooms. (figure.ai)
