Figure AI says its Figure 03 humanoid robot system has completed a 200-hour autonomous package-sorting run. The test started as an eight-hour livestreamed demonstration at the company’s San Jose headquarters, but was extended after the robots continued operating without what Figure described as a failure. According to Figure CEO Brett Adcock, the run was stopped after reaching 200 hours. The company presented the test as a demonstration of reliability and endurance for humanoid robots in a logistics workflow.
The task shown during the livestream was narrowly defined. The robots had to identify packages, pick them up and place them on a conveyor belt with the barcode facing down. Figure said the system operated autonomously using its Helix-02 AI model. The company has stated that the robots were not teleoperated during the test, and that automatic reset behavior was used when the system encountered situations outside its normal operating range.
The test used multiple Figure 03 units rather than a single robot operating continuously for the full duration. During the livestream, robots rotated in and out of the work cell, with some units charging while others continued sorting. This setup is relevant for warehouse applications, where uptime depends not only on the performance of one robot but also on fleet coordination, charging, recovery and handover between units. Earlier in the run, Figure reported that the system had passed the 24-hour mark while sorting tens of thousands of packages. Adcock also posted that the robots had crossed 50 hours of continuous operation and had sorted more than 63,000 packages at that stage. Independent reporting on the livestream described the setup as a repetitive but measurable logistics task, with the robots operating close to human sorting speed in that specific scenario.
The 200-hour claim follows a separate 10-hour comparison between a Figure humanoid and a human intern. In that test, the human worker sorted 12,924 packages, while the robot sorted 12,732, a difference of 192 packages. The reported average times were 2.79 seconds per item for the human and 2.83 seconds for the robot. The comparison gave Figure a public benchmark for speed on this particular task, but it did not test broader warehouse activities such as exception handling, mixed workflows or operation in less controlled environments.
The test does not mean that Figure 03 is ready for general-purpose warehouse deployment. The work cell was designed around one repeated task, with known object types and a controlled conveyor setup. Several observers also noted that the livestream showed occasional errors and pauses, including questions about package orientation and recovery behavior. Business Insider quoted roboticist Ayanna Howard as saying the performance showed progress, but also pointed to accuracy issues that would matter in commercial operation.
Still, long-duration operation is an important issue for humanoid robotics. Many public robot demonstrations focus on short tasks or edited clips. A multi-day run, even in a constrained environment, provides information about uptime, recovery, battery management, hardware wear and the ability to rotate robots through a work process. For warehouses and factories, these factors are often as important as peak manipulation speed.
We just wrapped what began as an 8-hour challenge – and it ran for 200 hours without a failure
Shoutout to the team for the hardcore engineering behind F.03 and the robust Helix models powering it pic.twitter.com/ir47UasJOn
— Brett Adcock (@adcock_brett) May 22, 2026
