Deployment maps are only useful when each entry states the robot generation, site, task, date, evidence source and whether the record is a pilot, commercial deployment or public demonstration.
Original X post
Open on XLoading the full X post…
Key facts
- Figure reports that Figure 02 ran more than 1,250 hours at BMW Plant Spartanburg, loaded more than 90,000 parts and contributed to more than 30,000 X3 vehicles.
- Figure 03 returned to BMW Plant Spartanburg in 2026 for a logistics sequencing workflow, which is a separate robot generation and task from the Figure 02 record.
- Agility reports more than 65,000 operating hours for Digit 4 across customer sites and a 100,000-tote milestone at GXO at about 98% on-task accuracy.
A map can mislead if it treats every video as deployment
A pin on a map can make a staged demonstration look equivalent to a customer deployment. Each entry should state the robot generation, named site, task, date, control boundary and source before it is counted as deployment evidence.
A factory pilot, a customer trial, a research pilot and a conference demo do not mean the same thing. They have different stakes, different supervision levels and different evidence quality. A good map should show those layers instead of flattening them into one headline.
Documented pilots are not the same as scale
A pilot can be real while remaining narrow. One site, one task and substantial engineering support can produce useful operating evidence without establishing fleet-wide maturity.
Current public records show why generation labels matter. Figure 02 has a published BMW production record, while Figure 03 later returned for a different sequencing workflow. Agility's long-running operating metrics refer to Digit 4, while Digit 5 was announced in September 2026 for later availability. A map should keep those records separate.
What a better humanoid map should include
Useful fields include robot generation, company, country, named site, task, deployment status, source date, autonomy boundary, unit count when confirmed, operating hours, intervention rules and whether the metric is company-reported or independently documented.
When a source omits unit count, intervention rate or failure distribution, the map should leave that field undisclosed rather than infer it from a video or announcement.
Sources and methodology
Share this article
Share the current TechniaHQRobot article page.
Continue reading
Open the latest robotics reporting, Physical AI analysis and hardware notes.