luU57hMhkak · ↓ 11 archive videos & press below
@figureai →
What we know
What we don't know
Open questions (post-stream)
Final status at stream end (20 May 2026, 01:37 UTC)
Performance Over Stream Time
Key Insights
| Window | Rate (pach/h) | vs Overall |
|---|
| Day / hours | Rate | Packages |
|---|
Robot Shifts Timeline (continuous-run durations)
Stream Segments — What Happened When
Achievements Unlocked
Package Mix · final observation (20 May 2026)
By colour
By size
Avg in frame
Event Feed · last 30 semantic observations (final hours, 21–22 May 2026)
Robot Errors Observed (13–20 May 2026)
External Signals — News & Discussion (during 13–20 May 2026)
OCR Cost · final spend across tracking window
Patterns & Predictions
Robot Roster
| Robot | Model | Rate | Packages | Hours | Sightings |
|---|
Human vs Robot — F.03 Challenge
F.03 robot fleet (Bob/Frank/Gary/Rose/Jim, rotating): 12,732 packages · ~1,273/h
Margin: Aimé won by 192 packages (≈1.5%). F.03 led briefly at hour ~5 when Aimé took a bathroom break, but Aimé retook the lead and held it to the bell. Aimé reported blistered fingers and a forearm that felt "basically broken."
⚠ The numbers below were the dashboard's *previous* live framing. They compare a hardcoded "1,431/h human baseline" — which does not match published industry data. Real sort-to-destination throughput for an Amazon-class worker is 60–400 pkg/h (peak); 1,000+/h numbers in industry literature are *single-station induct feeds*, not full sort-to-bin. By that honest benchmark, F.03 sustained 1,248/h was already 3-20× a typical human sorter — Aimé's 1,292/h was an outlier peak-induct performance, not a normal human baseline. Prime Vision case study
Probable Charging Dock Events
About this site & disclaimers
⚠ Read before sharing or quoting
This page is an independent fan-built monitor of the public Figure AI humanoid-robot livestream on YouTube. It is not affiliated with, endorsed by, or operated by Figure AI Inc. The video stream is the property of Figure AI and is embedded here only via YouTube's standard public-embed mechanism; we do not redistribute, record, or relay the video itself.
All numbers, rates, trends, segment narratives, time-of-day patterns, predictions, and
verdicts shown above are produced by an autonomous AI vision pipeline
(currently gpt-5.4-nano reading single frames every
~60 seconds) plus deterministic post-processing. Vision OCR can mis-read digits, mis-identify
robots, or skip frames. Calculations are estimates for entertainment and educational
purposes only and should not be used for any commercial, financial, contractual,
journalistic, or operational decision.
Continuous development
The site is in permanent active development. Features change daily; old verdicts may be revised as the pipeline improves. The exact formulas and source code evolve — what you see today may differ tomorrow.
Calculations may be wrong
OCR mis-reads happen. Sanity validators reject obviously-bad readings but subtle errors can slip through. Every metric carries implicit ±5–15% uncertainty; confidence levels are surfaced on the verdict card and the audit page.
Autonomous AI processing
Vision frames are sent to an OpenAI-compatible LLM gateway. Token usage and cost are tracked transparently in the "OCR Cost Tracking" panel. No human reviews each reading; the pipeline runs unattended.
Not affiliated · not financial advice
This project is a personal technical exercise. Nothing here is investment advice, a recommendation, a benchmark for any commercial product, or a measurement Figure AI Inc. would necessarily agree with.
Data & privacy
The site stores only what's visible on the public livestream (package counts, robot identities visible on chest stickers, timer readings). No personal information about viewers is collected. The dashboard has no analytics scripts beyond what YouTube embeds.
Trademarks
"Figure" and "F.03" are trademarks of Figure AI Inc. They are used here for descriptive and fair-use purposes only. All other names (ROSE, GARY, JIM, BOB, FRANK …) are identifiers we observed on the public stream.