Figure AI reveals Figure 03 humanoid robot

Knitwear over carbon fiber

For years, humanoid robotics had a distinct look: exposed cabling, polished metal, and the unmistakable vibe of a walking manufacturing tool. Figure just decided that era is over.

Figure officially revealed Figure 03, its third-generation humanoid robot. The standout visual is immediate: it comes dressed in tailored, machine-washable knitwear, modeling five different outfits in teaser footage like a runway model built for domestic chores.

Beneath the fabric, this is a wholesale mechanical and perceptual overhaul. Figure shaved 9% off the total weight compared to Figure 02 and visibly slimmed down the frame. Pinch points are padded with multi-density foam, and the torso houses a battery pack certified to UN38.3 standards, delivering up to five hours of runtime.

A paperclip's worth of touch

The real breakthrough lives in how Figure 03 perceives what it touches. Off-the-shelf tactile sensors were too fragile for constant physical labor, so Figure developed its own first-generation tactile sensor in-house.

The new fingertips can register forces as feather-light as three grams—essentially the weight of a paperclip resting on a finger. That level of sensitivity lets Figure's Helix VLA model detect the exact millisecond an object begins to slip, adjusting grip force dynamically before anything drops.

Vision received an equally aggressive refresh. The head cameras deliver double the frame rate, one-quarter the latency, and a 60% wider field of view. Figure also embedded low-latency wide-angle cameras directly into the palms of the hands. If the bot reaches deep into a low cabinet where its torso vision is completely blocked, the hand itself maintains eyes on the target.

When power runs low, Figure 03 walks over to a wireless charging pad. Induction coils built into its feet pull 2 kW of power without human intervention, while 10 Gbps mmWave links offload terabytes of real-world interaction data to train the next iteration of Helix.

Stamping out 100,000 units

While Figure 02 was heavily reliant on CNC machining—great for prototypes, terrible for unit economics—Figure 03 shifts decisively toward high-volume automotive techniques like die-casting, injection molding, and metal stamping.

To make that bet work, Figure unveiled BotQ, an in-house manufacturing plant designed specifically for high-rate humanoid assembly. BotQ's initial line is built to produce up to 12,000 bots per year, with a four-year roadmap targeting 100,000 completed units. Critical assemblies like actuators, battery enclosures, and sensor packages are kept completely inside Figure's supply chain.

This isn't purely a domestic experiment either. Figure 02 has already logged five straight months of ten-hour shifts assembling BMW X3s. Figure 03 pushes actuator speeds by 2x with higher torque density, specifically tailored to accelerate pick-and-place cycle times in warehouse settings.

Why it matters

We think Figure 03 marks the clearest inflection point yet from robotics as bespoke research to robotics as an appliance. By prioritizing washable soft goods and foot-based wireless docking, Figure is designing around the messy realities of lived-in spaces rather than controlled showroom stages.

For the broader industry, Figure's playbook puts uncomfortable pressure on competitors like Tesla's Optimus and 1X. Developing a capable vision-language-action model is no longer enough; hardware survivability and tooling economics are now table stakes. Committing to custom stamping and in-house component fabs is an expensive, high-stakes flex that few startups can afford.

The lingering question is hardware lock-in. Tooling for die-casting costs millions upfront and sharply restricts design agility. If real-world deployments over the next year reveal fundamental blind spots in joint kinematics or sensor placement, Figure will have to swallow massive retooling penalties to fix them.