Humanoid trial runs go public — Figure 03 at Amazon and Apollo V2 at Mercedes are the first paid pilots that survived contact with real warehouses
Figure announced an Amazon warehouse pilot for Figure 03 the same week Apptronik confirmed Apollo V2 in a Mercedes assembly cell. Both are paid pilots running in real production environments, not staged demos. The humanoid robotics category just crossed the line from demo-tier to operational-trial.
What landed
Figure 03 entered a paid pilot at an Amazon fulfillment center doing tote handling, pick-and-place, and shelf restock alongside human workers. Apptronik Apollo V2 is in a Mercedes assembly cell doing torque-tool handoff and small-part subassembly.
The 'real warehouse' threshold
For two years the humanoid robotics category has been dominated by staged demos. Boston Dynamics Atlas parkour. Figure folding laundry. Tesla Optimus walking. Those were marketing artifacts, not production-pilot data. Figure 03 at Amazon and Apollo V2 at Mercedes are the first two pilots where the robot is paid to handle real product on a real timeline, and where the pilot can fail in measurable ways.
Why the failure modes matter
The interesting data from these pilots isn't the success rate — it's the failure mode distribution. Where does the robot get stuck? What kinds of tasks need human intervention? How does throughput compare to a human worker, adjusted for the robot's 24/7 uptime? Those answers determine whether the unit economics work, and the pilots are sized to produce that data over 90-180 days.
Humanoid robotics just crossed from demo-tier to operational-trial. The next 6 months will produce the first credible economic data on per-robot-hour productivity in real production environments.
The competitive backdrop
Tesla Optimus Gen 3 is still in pilot-production at Fremont, not yet shipping to external customers. 1X has consumer preorders but no commercial pilot at this scale. Boston Dynamics is in industrial deployment with Spot/Stretch but not with Atlas. Figure and Apptronik just opened a competitive window where they're the only humanoid platforms in paid production pilots — that window probably closes in 6-12 months as Tesla and 1X catch up.
The forward read
- One pilot produces unit-economics data that justifies fleet expansion by Q4. That sets the per-robot-hour rate the rest of the market negotiates against.
- The first humanoid-robotics IPO files by H1 2027. Figure or Apptronik is the most likely; pilot data is the prerequisite.
- Labor-impact conversations move from theoretical to operational. Once the per-robot-hour rate is public, the wage-comparison conversation becomes concrete.
Reuters — Figure 03 Amazon warehouse pilot → · Bloomberg — Apptronik Apollo V2 Mercedes deployment → · IEEE Spectrum — Humanoid pilot economics analysis →