Figure AI 200-Hour Sorting Livestream: A Humanoid Robot Industrial Milestone
On May 14, at a logistics warehouse in California, three Figure 03 humanoid robots named Bob, Frank, and Gary stepped onto the sorting line.
There was no power switch. As Figure AI CEO Brett Adcock put it: “This livestream has no off button.”
An originally planned 8-hour validation test turned into an 8-day, 200-hour uninterrupted industrial endurance marathon. The three robots collectively sorted 249,558 packages, averaging 2.88 seconds per package, with zero major hardware failures throughout.
This was not a product launch. It was a public on-the-job assessment.
Core Livestream Data
| Metric | Data |
|---|---|
| Livestream Duration | 200 hours (May 14–May 22) |
| Total Packages Sorted | 249,558 |
| Average Processing Speed | ~2.88 seconds per package |
| Number of Robots | 3 (rotational shift system) |
| Single Robot Battery Life | 3–4 hours |
| Major Failures | Zero |
The three robots operated on a rotational schedule: each worked until its battery dropped to 20%, then autonomously requested a teammate to take over and proceeded to the charging station. The entire process required no human intervention.
Human vs. Machine: Humans Won by 0.04 Seconds, But Couldn’t Continue
On May 19, Figure AI arranged a 10-hour human-robot side-by-side competition.
Results:
- Human worker: Sorted 12,924 packages, average 2.79 seconds per package
- Figure 03: Sorted 12,732 packages, average 2.83 seconds per package
Humans led by only 192 packages in total, with a per-package speed advantage of 0.04 seconds.
But after the match, the human worker had blisters on their fingers and a left arm too sore to lift. The three robots, meanwhile, swapped batteries and kept sorting.
Brett Adcock’s post-match statement: “This may be the last time humans beat robots at a sorting task.”
Helix-02: Industrial Validation of End-to-End Neural Networks
The brain of Figure 03 is Helix-02, an end-to-end full-body autonomous neural network system.
Unlike traditional industrial robots with layered control, Helix-02 is a unified single model:
- High-frequency low-level controller: Directly drives joints
- Visuomotor policy layer: Maps camera pixels to action commands
- Semantic reasoning layer: Processes language instructions and scene understanding
All inference runs on-device, with no cloud dependency. The robot “reasons directly from camera pixels.”
This proves the feasibility and stability of end-to-end embodied intelligence solutions in industrial scenarios.
Production Speed: One Robot Per Hour
During the livestream, Brett Adcock revealed a key number: Figure’s BotQ factory currently produces one F.03 robot per hour.
120 days ago, that number was one per day. Production efficiency has scaled 24x.
Over 55 units were produced in the livestream week alone. The F.04 design is fully locked, and component delivery processes have already started.
Adcock called F.04 “the biggest leap between robot generations.”
Commercial Validation: Cost Down 62%
Figure AI has partnered with DHL and two other global logistics giants to advance deployment.
Official data shows: the first batch of 20 units ran continuously for 30 days at a California e-commerce warehouse, reducing picking costs by 62% and cutting labor needs by over 50%.
The average hourly wage in U.S. warehousing and logistics is about $20. One robot can run 2–3 shifts for 24 hours, with electricity costs nearly negligible. The effective “wage” may be less than half of human labor.
And it never calls in sick, never gets injured, and never has an attitude.
Controversies and Limitations
The most questioned moment during the livestream: the F.03 suddenly raised its hand and touched its own head.
The comment section exploded: was this autonomous behavior, or remote control?
Figure explained that the robot was avoiding an iron frame structure beside the conveyor belt. But some viewers captured frames from other timestamps, arguing there was no visible obstacle nearby.
Industry insiders also pointed out the limitations of the test environment:
- Package specifications were highly uniform (only colors differed; sizes and thicknesses were identical)
- Real-world courier SKU complexity far exceeds the test scenario (irregular shapes, fragile items, liquid packages, etc.)
- The livestream mostly featured easy-to-handle soft packs; actual irregular cardboard boxes are significantly harder
One embodied intelligence practitioner commented: “Figure 03 is indeed industry-leading in continuous operational stability and motion fluidity, but the test scenario was too idealized to be broadly representative.”
Industry Signals
This livestream delivered several clear signals:
- Humanoid robot industrial endurance has been validated: 200 hours of uninterrupted operation is not a demo edit—it’s a real industrial scenario
- End-to-end neural network solutions are viable: Helix-02 proved stability and adaptability in complex environments
- The mass production inflection point is near: One robot per hour means scaled deployment is becoming feasible
- Logistics is the first blue-collar job to be disrupted: High U.S. blue-collar labor costs make the ROI calculation straightforward
Figure AI is not putting on a marketing show. It is proving one thing to enterprise customers and investors: this thing can work as an employee.
Sources
- National Business Daily, May 22, 2026
- Science and Technology Innovation Board Daily, May 15, 2026
- IT Home / Sina Tech, May 24, 2026
- Figure AI official social media livestream and Brett Adcock tweets