Figure scales humanoid output 24x

Humanoid robots have had a scaling problem for years. Startups could show a polished demo, maybe even a pilot, but turning that into repeatable factory output was the part that stayed fuzzy. Figure is trying to close that gap now. On April 29, the company said its BotQ factory ramped Figure 03 production from one robot per day to one per hour in under 120 days, with more than 350 robots delivered so far. (figure.ai) ### Why is one robot per hour a big deal? Because this is the first number that sounds like a factory, not a lab. Figure says BotQ has moved from prototype work into dedicated production lines for the robot’s critical modules, and that the one-per-hour cycle time is what it needs for its current Figure 03 targets. That does not mean the li(figure.ai)does mean the company is claiming a repeatable manufacturing cadence instead of one-off assembly. (figure.ai) ### What changed inside the factory? Basically, Figure rebuilt the process around volume. Earlier BotQ plans centered on in-house manufacturing, factory software, and a redesign away from slow prototype-style CNC machining toward injection molding, die casting, metal injection molding, and stamping. Figure says the current ramp was powered(figure.ai)ed workstations, plus tighter supplier qualification and more than 50 in-process inspection points. (figure.ai) ### Are the robots actually coming off the line cleanly? That is the real question — and Figure’s answer is yield. The company says end-of-line first-pass yield is now above 80% and improving weekly. It also says the battery line hit 99.3% first-pass yield, shipped more than 500 battery packs, and produced more than 9,000 actuators across 10-plus SKUs. Each robot(figure.ai)ification tests before sign-off, plus burn-in sessions with repeated full-body motions like squats and jogging. (figure.ai) ### What are those 350 robots doing? Not all of them are going straight to paying customers. Figure says shipped robots are being allocated across internal R&D, data collection, home-task development, and commercial use-case work. That matters because a bigger fleet does two jobs at once — it tests whether manufacturing is real, and it gen(figure.ai), more hardware is supposed to make the software better. (figure.ai) ### Where does Helix fit into this? Helix is Figure’s in-house vision-language-action system — the software layer that handles perception, movement, and reasoning in real time on Figure 03. Figure built Figure 03 around Helix, with a redesigned sensor suite, new hand system, wider-field cameras, lower latency, and embedded palm cameras for(figure.ai)tonomy progress, the claim is that more robots in the field create the training and test loop Helix needs. (figure.ai) ### Is this the same as commercial breakout? Not yet. The catch is that these numbers are still company claims, and “delivered” here includes internal and development allocation, not just broad outside deployment. Figure also said back in March 2025 that BotQ’s first-generation line was designed for up to 12,000 humanoids per year, so the bigger story is not the theo(figure.ai)y can hold yield, reliability, and cost while climbing toward it. (figure.ai) ### Why does this matter beyond Figure? Because the humanoid race has shifted. The hard part is no longer just getting a robot to walk, grasp, or climb a ramp in a video. The hard part is building lots of them with stable quality, then using that fleet to improve autonomy fast enough to justify the hardware spend. Figure is arguing that those two loops — manufact(figure.ai)her. (figure.ai) ### Bottom line The headline is simple: Figure says it can now build humanoids at a real factory cadence. If the yield holds and those robots start proving useful outside internal programs, this stops looking like a robotics demo story and starts looking like an industrialization story. (figure.ai)