Figure AI’s 200‑Hour Humanoid Robot Stress Test Explained

However, not all reporting agrees: one source cited 149,000+ packages, highlighting some uncertainty in the exact totals published during the livestream.

Even with that discrepancy, the demonstration still showed that the robots could run for more than a week of continuous warehouse‑style work without teleoperation or mechanical breakdowns—an important milestone for real deployments.

How Fast the Robots Worked

Throughput was the main benchmark for the test.

Figure AI previously reported that its logistics system had reached “human parity” speeds, processing packages at under three seconds per item—roughly the pace of warehouse workers performing the same task during a shift.

The workflow in the demo involved several steps typical in parcel‑sorting operations:

• visually identifying the package
• orienting it so the barcode can be scanned
• scanning or verifying the label
• placing the item onto the correct conveyor line

Maintaining a pace of about 3 seconds per package across hundreds of thousands of items suggests the robots can sustain production‑grade throughput rather than short bursts of speed.

The AI Brain Behind the Robots: Helix‑02

The performance of the F.03 robots depends heavily on Figure’s in‑house AI system, Helix‑02.

Helix‑02 is described as a vision‑language‑action (VLA) model—a neural system that merges perception, reasoning, and motor control into a single architecture. Instead of separate software modules, one learned policy interprets camera input and produces the robot’s full‑body actions.

This architecture enables the robot to:

• perceive objects using cameras and sensors
• interpret tasks or instructions
• plan and execute movements with its arms, hands, and body

The system effectively converts “pixels to actions”, allowing the robot to manipulate unfamiliar packages and adapt to changing conditions in real environments.

For logistics tasks like package sorting, that means recognizing different package shapes—boxes, envelopes, or poly bags—and adjusting how the robot grips or flips them to scan labels correctly.

The “Man vs. Machine” Contest

To benchmark the robots against humans, Figure staged a 10‑hour head‑to‑head contest between one F.03 robot and a company intern named Aime.

Both were assigned the same task: scan each parcel and place it correctly on a conveyor belt.

The final results were extremely close:

• Human intern: 12,924 packages (≈ 2.79 seconds per package)
• F.03 robot: 12,732 packages (≈ 2.83 seconds per package)

The human won by just 192 packages, a gap of roughly 1.3% over the entire shift.

In other words, the robot worked at almost exactly the same pace—but not quite enough to beat the human in that single shift.

Why the Test Matters for Warehouse Automation

Humanoid robots have existed for years, but most demonstrations historically lasted minutes or hours—not days.

The Figure experiment highlights three changes that could matter for logistics automation:

1. Endurance
Running for 200 hours continuously shows the hardware and control systems can handle long operational periods, which is essential for real warehouses.

2. Near‑human throughput
Sorting packages in roughly three seconds each places the robots close to human productivity on repetitive tasks.

3. Autonomous operation
The robots reportedly operated without teleoperation, meaning the AI system handled perception and manipulation directly on the robot.

Together, these factors suggest humanoids may be moving from short demonstrations toward commercially relevant warehouse tasks.

The Important Caveat

Despite the impressive numbers, the evidence still comes largely from Figure’s own demonstrations and livestreamed tests, rather than independent large‑scale deployments.

That means the experiment is best understood as a proof of progress rather than proof that humanoid robots are ready to replace large portions of warehouse labor today.

Still, the narrow margin in the “Man vs. Machine” contest—combined with a week‑long autonomous run—signals how quickly humanoid robotics is advancing toward real industrial work.

The intern may have won this round, but the gap between humans and humanoid robots in logistics tasks is now measured in seconds per package, not orders of magnitude.