On May 18, 2026, Figure AI CEO Brett Adcock published the final results of an unusual experiment: a 10-hour head-to-head sorting challenge between human intern Aime and the company's fleet of autonomous Figure 03 humanoid robots. The human won — but barely, and only for the duration of the shift.
Key takeaways
- Human intern Aime: 12,924 packages at 2.79 sec/package average
- Figure 03 fleet: 12,732 packages at 2.83 sec/package average
- After the 10-hour shift, the robot fleet kept running — surpassing 116 hours of continuous operation and 145,000+ packages sorted
- The intern finished with blistered fingers and a "basically broken" left forearm from repetitive strain
- Robots ran on Figure's proprietary Helix-02 end-to-end vision-language-action neural network
Rules of the challenge
Both sides performed the identical task: detect the barcode on each package, pick it up, and place it face-down on a moving conveyor belt. The challenge ran for 10 hours — May 17, 10 AM to 8 PM PT. Per California labor law, the intern received mandatory meal and rest breaks totaling roughly 50 minutes. The robot fleet rotated units (Bob, Frank, Gary, Rose, Jim) seamlessly when batteries needed charging, with zero downtime.
Helix-02: the technology inside Figure 03
Figure 03 runs on Helix-02, an end-to-end vision-language-action (VLA) model that processes raw camera pixels directly into actuator commands. There are no separate perception or planning modules — the model takes image input and outputs servo motor vectors in a single forward pass.
Competitive context
Agility Robotics' Digit is testing similar logistics tasks for Amazon, but focuses on bin transport rather than individual item sorting. Boston Dynamics has not disclosed comparable throughput data for Atlas in industrial settings. Tesla Optimus has no publicly verified sorting benchmark. Figure is therefore the first company to publish a direct human-vs-robot comparison with live, unedited footage in near-production conditions.
Why does it matter?
The automation debate usually operates on projections and anecdotes. Figure delivered something rare: hard throughput data from a real logistics environment, broadcast live without editing. The human win in the sprint is less important than the marathon effect: a fleet of five robots, running continuously for over a week and sorting more than 145,000 packages, makes the economic case for automation — 24/7 availability, no wages, no fatigue, no sick days — measurable and verifiable for the first time at this scale.
What's next?
- Figure 04 is in design lock — Adcock called it the "iPhone 1 Moment" for the company; it targets improved speed and lower unit cost
- The livestream continues — Adcock has not announced an end date; the fleet has already run for 116+ hours
- Future Man vs. Machine challenges expected when Figure 04 ships — Adcock stated the human win on May 17 was the last one
