US Robotics player Figure has rapidly scaled production of its Figure 03 humanoid robots, marking a shift from prototype to mass manufacturing.

At its manufacturing facility, BotQ, in California, the company boosted output from one robot per day to one per hour—a 24-fold increase achieved in under four months.

The ramp-up, supported by custom software and over 150 networked workstations, has enabled the delivery of more than 350 units.

According to Figure, with improved supplier quality and rigorous inspection processes, production yields are rising steadily. The expansion is also accelerating data generation, a key factor in advancing the robots’ autonomous capabilities.

Last year, Figure revealed its plans for the BotQ facility, an in-house factory dedicated to humanoid robots, targeting 12,000 units annually with plans to scale production further.

Humanoid output scales

Figure has transitioned its BotQ facility from prototype development to full-scale production, establishing dedicated lines for all critical modules of the Figure 03 humanoid robot.

The company has achieved a one-unit-per-hour production cycle, supported by custom manufacturing software operating across more than 150 networked workstations. To improve yield and efficiency, it strengthened supplier quality through strict qualification processes and implemented over 50 in-process inspection points.

As a result, end-of-line first-pass yield has exceeded 80 percent, while battery production has reached a 99.3 percent yield, with over 500 units shipped and more than 9,000 actuators produced.

Each robot undergoes over 80 functional tests, including stress and burn-in exercises such as squats and jogging, to eliminate early failures. The growing fleet is accelerating development by generating large-scale data for Helix, the company’s humanoid AI model, while enabling real-world deployment across research, commercial, and domestic use cases, according to a blog by Figure.

Operational scale has improved diagnostics, enabling rapid failure analysis, while software fallback systems allow robots to maintain functionality during minor faults. With high-frequency issues resolved, focus has shifted to edge-case failures. To support expansion, Figure has built internal service, fleet management, and over-the-air update systems, ensuring continuous monitoring, upgrades, and feedback integration across all deployed robots.

Perception meets motion

Figure has also introduced a major upgrade to its AI model Helix System 0 (S0), enabling perception-conditioned whole-body control for its humanoid robots. P

Previously, the model relied solely on proprioception—understanding joint states and body motion—limiting its ability to navigate complex environments like stairs or uneven terrain. With the latest update, S0 now integrates visual input from onboard stereo cameras, converting RGB images into a 3D spatial representation of the surroundings. This allows the robot to both “feel” and “see” the terrain it navigates.

The system is trained end-to-end using reinforcement learning in simulation across diverse, randomized terrains. Notably, the learned behaviors transfer directly to real-world environments without additional calibration or fine-tuning, overcoming a long-standing sim-to-real challenge. Robots can now traverse stairs and varied surfaces with human-like stability and motion, even under changing lighting conditions.

According to Figure, the advancement marks a significant step toward fully autonomous, perception-driven control systems. Beyond stair navigation, the architecture supports a wide range of environment-aware behaviors. Combined with high-volume production at BotQ, Figure is rapidly scaling its robot fleet, accelerating data collection, improving reliability, and strengthening its position in the race to deploy capable humanoid robots at scale.