Atlas of the Present Atlas · v0.1

As of 12 September 2026

Dossier 08

Robotics

v1.5-draft · unchecked

1. As of

Date
Version
v1.5-draft
Author / model
Atlas generator
Reviewer
unchecked (Josef)

2. In one sentence

What ships is industrial robots (542,076 installations in 2024, stock 4.66 million) and vacuums/AMRs — not humanoids; lab dexterity (Rubik/DeXtreme), RT-1/RT-2 kitchen manip (EDR; VLA unseen ~62%), ALOHA/ACT, OpenVLA/OXE and π0 (Black et al. 2024, E1 architecture: PaliGemma + flow-matching action expert, >~10,000 h, 7 configurations / 68 tasks) are lab measurements — far from factory routine and household autonomy; vendor humanoid pages are demos with their own caveats.

Established now · E1 / E2

3. What works today

Industrial robots: what ships

  1. E1

    542,076 industrial robot installations in 2024; stock 4,663,698. IFR method over 12 years, not a census. China 295,045 = 54%. Electronics 24% overtook auto 23%. Density 2024 world 177 per 10,000 employees.

    IFR WR 2025 Exec. World Robotics 2025 — Industrial Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Industrial_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.IFR press 2025-09-25. Global Robot Demand in Factories Doubled in 10 Years. https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years. As of 2025-09-25. Checked 2026-08-28. Type: Association press release.IFR Sources. World Robotics 2025 — Sources and Methods. https://ifr.org/img/worldrobotics/Sources___Methods_WR_2025_Industrial_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.
  2. E1

    Density press release 2024-11-20 refers to 2023 data: world 162, Korea 1,012. Do not mix with 2024 density 177.

    IFR density 2024. Global Robot Density in Factories Doubled in Seven Years. https://ifr.org/ifr-press-releases/news/global-robot-density-in-factories-doubled-in-seven-years. As of 2024-11-20 (2023 data). Checked 2026-08-28. Type: Association press release.
  3. E1

    ISO 8373 via IFR: teleoperation is not a robot. AMR without an arm = service, not industrial. Professional service sample >199,000 (not extrapolated); consumer ~20.1 million, 97% domestic; care-at-home 536.

    IFR Sources. World Robotics 2025 — Sources and Methods. https://ifr.org/img/worldrobotics/Sources___Methods_WR_2025_Industrial_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.IFR Service Exec. World Robotics 2025 — Service Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Service_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.
  4. E2

    Humanoids: “no massive use today” (IFR slides). Infographic: humanoids will not compete with industrial robots on speed, precision, reliability.

    IFR SR slides. World Robotics 2025 — Service Robots Presentation. https://ifr.org/downloads/press_docs/Press_Conference_2025_SR.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.IFR humanoid press. Humanoid Robots: Vision and Reality Paper Published by IFR. https://ifr.org/ifr-press-releases/news/humanoid-robots-vision-and-reality-paper-published-by-ifr. As of 2025-08-14. Checked 2026-08-28. Type: Association press release.IFR humanoid info. Humanoid Robots — Vision and Reality (Infographic). https://ifr.org/downloads/press_docs/Humanoids_Position_Infograph_2025.pdf. As of 2025-08. Checked 2026-08-28. Type: Association PDF.

Lab dexterity, measured

  1. E1

    OpenAI Rubik (Table 6): full scramble 20% with Giiker cube, 0% vision-only. N=10. Cage, markers. Not 100% autonomous cubing.

    OpenAI Rubik. Solving Rubik's Cube with a Robot Hand. https://arxiv.org/html/1910.07113. As of 2019-10. Checked 2026-08-28. Type: Paper.
  2. E1

    DeXtreme: real mean ~28 consecutive; cables burn; seed variance 1–112. Sim-to-real transfer exists and is unstable.

    DeXtreme. DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality. https://arxiv.org/html/2210.13702v2. As of 2022-10. Checked 2026-08-28. Type: Paper.
  3. E1

    Reality gap is a construct with a taxonomy, not a solved engineering problem. Aljalbout et al. 2025: the gap remains the central transfer diagnosis.

    Reality Gap. The Reality Gap in Robotics: Challenges, Solutions, and Best Practices. https://arxiv.org/html/2510.20808. As of 2025-10. Checked 2026-08-28. Type: Paper (survey).

Fine manipulation ALOHA/ACT

  1. E1

    ALOHA (Zhao et al., arXiv:2304.13705): low-cost bimanual teleop system <$20k. ACT = Action Chunking with Transformers (CVAE); ~80M params; 50 Hz; chunked actions + temporal ensembling. 50 human demos/task (~10 min), Thread Velcro 100.

    Zhao et al. ALOHA/ACT. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. https://arxiv.org/pdf/2304.13705. As of 2023-04. Checked 2026-09-08. Type: Paper (arXiv).
  2. E1

    Real final success (Table I/II, human data): Slide Ziploc 88%; Slot Battery 96%; Open Cup 84%; Put On Shoe 92%; Prep Tape 64%; Thread Velcro 20%.

    Zhao et al. ALOHA/ACT. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. https://arxiv.org/pdf/2304.13705. As of 2023-04. Checked 2026-09-08. Type: Paper (arXiv).
  3. E1

    Baselines BC-ConvMLP / BeT / RT-1 / VINN near 0 final success on Slot Battery / Slide Ziploc (paper).

    Zhao et al. ALOHA/ACT. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. https://arxiv.org/pdf/2304.13705. As of 2023-04. Checked 2026-09-08. Type: Paper (arXiv).

VLA and X-embodiment, lab measurement

  1. E1

    Open X-Embodiment (arXiv:2310.08864, body opened): dataset from 22 robots / 21 institutions; 527 skills / 160,266 tasks (abstract/Fig. 1).

    OXE / RT-X. Open X-Embodiment: Robotic Learning Datasets and RT-X Models. https://arxiv.org/pdf/2310.08864. As of 2023-10. Checked 2026-09-06. Type: Paper (arXiv).
  2. E1

    OXE Fig. 4: RT-1-X mean success rate 50% higher than Original Method or RT-1 on small-scale domains (same RT-1 vs RT-1-X architecture; co-training on the X-embodiment mixture).

    OXE / RT-X. Open X-Embodiment: Robotic Learning Datasets and RT-X Models. https://arxiv.org/pdf/2310.08864. As of 2023-10. Checked 2026-09-06. Type: Paper (arXiv).
  3. E1

    OXE Table I (large-scale Bridge/RT-1): RT-1-X can underfit (e.g. Bridge IRIS WidowX RT-1 40% vs RT-1-X 27%); RT-2-X 55B there 50%. Positive transfer needs capacity.

    OXE / RT-X. Open X-Embodiment: Robotic Learning Datasets and RT-X Models. https://arxiv.org/pdf/2310.08864. As of 2023-10. Checked 2026-09-06. Type: Paper (arXiv).
  4. E1

    OpenVLA (Kim et al., arXiv:2406.09246): 7B open VLA (Llama 2 + DINOv2/SigLIP), fine-tuned on 970k Open-X trajectories; +16.5 percentage points absolute vs RT-2-X 55B across 29 tasks (WidowX + Google Robot).

    Kim et al. OpenVLA. OpenVLA: An Open-Source Vision-Language-Action Model. https://arxiv.org/pdf/2406.09246. As of 2024-06. Checked 2026-09-06. Type: Paper (arXiv).
  5. E1

    OpenVLA eval scope in the paper: BridgeData V2 170 rollouts (17×10); Google Robot 60 rollouts (12×5). Training 21,500 A100-hours. Weights/code stated open — lab success ≠ factory humanoid.

    Kim et al. OpenVLA. OpenVLA: An Open-Source Vision-Language-Action Model. https://arxiv.org/pdf/2406.09246. As of 2024-06. Checked 2026-09-06. Type: Paper (arXiv).

RT-1: Robotics Transformer, lab kitchen

  1. E1

    RT-1 (Brohan et al., arXiv:2212.06817, body opened): ~130k episodes, 700+ tasks, 13 Everyday Robots mobile manipulators over 17 months. Architecture FiLM-EfficientNet + TokenLearner + Transformer; ~35M params; 3 Hz. Evaluation >3,000 real-world trials.

    Brohan et al. RT-1. RT-1: Robotics Transformer for Real-World Control at Scale. https://arxiv.org/pdf/2212.06817. As of 2022-12. Checked 2026-09-09. Type: Paper (arXiv).
  2. E1

    Table 2 success (%): RT-1 seen 97 / unseen 76 / distractors 83 / backgrounds 59; Gato 65/52/43/35; BC-Z 72/19/47/41; BC-Z XL 56/43/23/35. Abstract: +25 / +36 / +18 percentage points vs next-best baseline (new tasks / distractors / backgrounds).

    Brohan et al. RT-1. RT-1: Robotics Transformer for Real-World Control at Scale. https://arxiv.org/pdf/2212.06817. As of 2022-12. Checked 2026-09-09. Type: Paper (arXiv).
  3. E1

    Seen eval: >200 training instructions with varied placement/time/robot position. Unseen: 21 novel instruction combinations (skill×object partly in train, combo new).

    Brohan et al. RT-1. RT-1: Robotics Transformer for Real-World Control at Scale. https://arxiv.org/pdf/2212.06817. As of 2022-12. Checked 2026-09-09. Type: Paper (arXiv).

RT-2: Vision-Language-Action, web→robot

  1. E1

    RT-2 (Brohan et al., arXiv:2307.15818, body opened): VLA = co-fine-tune VLMs; actions as text tokens. Instantiations RT-2-PaLI-X (5B/55B) and RT-2-PaLM-E (12B). ~6,000 evaluation trials.

    Brohan et al. RT-2. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. https://arxiv.org/pdf/2307.15818. As of 2023-07. Checked 2026-09-10. Type: Paper (arXiv).
  2. E1

    Table 4 overall (%): Seen Tasks RT-2-PaLI-X-55B 91 / RT-2-PaLM-E-12B 93 vs RT-1 92 / MOO 75; Unseen Average both RT-2 62 vs RT-1 32 / MOO 35. Unseen Objects Easy/Hard: PaLI-X 70/62, PaLM-E 84/76 vs RT-1 31/43. Paper: ~2× generalization vs next baselines (RT-1/MOO).

    Brohan et al. RT-2. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. https://arxiv.org/pdf/2307.15818. As of 2023-07. Checked 2026-09-10. Type: Paper (arXiv).
  3. E1

    Table 5 emergent Average: RT-2-PaLI-X 60 vs RT-1 17 (Symbol Understanding avg 82 vs 16). Best model ~3× emergent vs RT-1; paper also “2x to 3x better than RT-1” on new instructions without additional robotic demos.

    Brohan et al. RT-2. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. https://arxiv.org/pdf/2307.15818. As of 2023-07. Checked 2026-09-10. Type: Paper (arXiv).

π0: VLA flow for cross-embodiment (Black et al.)

  1. E1

    π0 (Black et al., Physical Intelligence, arXiv:2410.24164 v4 Jan 2026; body opened 2026-09-12): VLM backbone initialized from PaliGemma; separate action expert with conditional flow matching for continuous action chunks; control up to ~50 Hz.

    Black et al. π0. π₀: A Vision-Language-Action Flow Model for General Robot Control. https://arxiv.org/pdf/2410.24164. As of arXiv 2024-10 (v4 Jan 2026). Checked 2026-09-12. Type: Paper (arXiv).
  2. E1

    Cross-embodiment pre-training: >~10,000 hours of robot data; pre-training on diverse data from 7 robot configurations and 68 tasks; mixture includes OXE Magic Soup subset, Bridge v2, DROID plus proprietary dexterous data; platforms: single/dual-arm and mobile manipulators (ALOHA-like).

    Black et al. π0. π₀: A Vision-Language-Action Flow Model for General Robot Control. https://arxiv.org/pdf/2410.24164. As of arXiv 2024-10 (v4 Jan 2026). Checked 2026-09-12. Type: Paper (arXiv).
  3. E1

    After fine-tuning, multi-stage dexterity: laundry folding (dryer→hamper→fold), table bussing/cleaning, box assembly; tasks can take minutes (paper: sometimes 5–20 minutes).

    Black et al. π0. π₀: A Vision-Language-Action Flow Model for General Robot Control. https://arxiv.org/pdf/2410.24164. As of arXiv 2024-10 (v4 Jan 2026). Checked 2026-09-12. Type: Paper (arXiv).
  4. E2

    Out-of-box / direct prompting: π0 outperforms OpenVLA and Octo (trained on the authors’ mixture) on reported out-of-box tasks; paper: “by far the best results… near perfect success” on shirt folding and easier bussing tasks — qualitative paper claim; exact % figures not quoted from OCR here.

    Black et al. π0. π₀: A Vision-Language-Action Flow Model for General Robot Control. https://arxiv.org/pdf/2410.24164. As of arXiv 2024-10 (v4 Jan 2026). Checked 2026-09-12. Type: Paper (arXiv).

Claimed · E3

4. What is claimed, not shown

IFR forecast, Figure, Atlas — vendor and association

  1. E3

    IFR forecast: 575,000 installations 2025, 700,000 2027. Forecast, not a measurement.

    IFR WR 2025 Exec. World Robotics 2025 — Industrial Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Industrial_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.
  2. E3

    Figure F.02: 90,000+ parts, 1,250+ hours, “contributed to 30,000+ X3” — vendor. KPI 84 s / >99% / zero interventions defined, not reported as achieved (E0 for achievement). Helix 02: 4-minute dishwasher, “results are early”. F.03 2026-06-30: product announcement.

    Figure F.02. F.02 Contributed to the Production of 30,000 Cars at BMW. https://www.figure.ai/news/production-at-bmw. As of 2025-11-19. Checked 2026-08-28. Type: Company report.Figure F.03. F.03 Arrives at BMW. https://www.figure.ai/news/f-03-at-bmw. As of 2026-06-30. Checked 2026-08-28. Type: Company report.Helix 02. Introducing Helix 02: Full-Body Autonomy. https://www.figure.ai/news/helix-02. As of 2026-01-27. Checked 2026-08-28. Type: Company report.
  3. E3

    Boston Dynamics Atlas CES 2026: 56 DoF, 50 kg, 2026 Hyundai/DeepMind — vendor. Same page: “currently in development” / prototype on stage. Electric Atlas 2024: end of hydraulics, not series factory use.

    Atlas CES. Boston Dynamics Unveils New Atlas Robot to Revolutionize Industry. https://bostondynamics.com/blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/. As of 2026-01-05. Checked 2026-08-28. Type: Company report.Electric Atlas. Introducing Electric Atlas. https://bostondynamics.com/news/introducing-electric-atlas/. As of 2024. Checked 2026-08-28. Type: Company report.

Constrained · Limit

5. Bottleneck and limit

What demo and forecast do not show

  1. E2

    Humanoids will not compete with industrial robots on speed/precision/reliability (IFR infographic). Battery ≠ full workday. ISO safety for legged robots “just started”.

    IFR humanoid press. Humanoid Robots: Vision and Reality Paper Published by IFR. https://ifr.org/ifr-press-releases/news/humanoid-robots-vision-and-reality-paper-published-by-ifr. As of 2025-08-14. Checked 2026-08-28. Type: Association press release.IFR humanoid info. Humanoid Robots — Vision and Reality (Infographic). https://ifr.org/downloads/press_docs/Humanoids_Position_Infograph_2025.pdf. As of 2025-08. Checked 2026-08-28. Type: Association PDF.
  2. E1

    Demo video ≠ rate. Rubik 20% / 0% vision; DeXtreme seed variance; reality gap. Teleop is not a robot per ISO.

    OpenAI Rubik. Solving Rubik's Cube with a Robot Hand. https://arxiv.org/html/1910.07113. As of 2019-10. Checked 2026-08-28. Type: Paper.DeXtreme. DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality. https://arxiv.org/html/2210.13702v2. As of 2022-10. Checked 2026-08-28. Type: Paper.Reality Gap. The Reality Gap in Robotics: Challenges, Solutions, and Best Practices. https://arxiv.org/html/2510.20808. As of 2025-10. Checked 2026-08-28. Type: Paper (survey).
  3. E1

    OpenVLA/OXE success rates are lab rollouts on named embodiments and authors’ protocols — not IFR installation statistics and not series humanoids on the factory floor.

    Kim et al. OpenVLA. OpenVLA: An Open-Source Vision-Language-Action Model. https://arxiv.org/pdf/2406.09246. As of 2024-06. Checked 2026-09-06. Type: Paper (arXiv).OXE / RT-X. Open X-Embodiment: Robotic Learning Datasets and RT-X Models. https://arxiv.org/pdf/2310.08864. As of 2023-10. Checked 2026-09-06. Type: Paper (arXiv).
  4. E1

    ALOHA/ACT lab success ≠ factory humanoid / IFR stock — imitation on a named low-cost setup and authors’ protocol.

    Zhao et al. ALOHA/ACT. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. https://arxiv.org/pdf/2304.13705. As of 2023-04. Checked 2026-09-08. Type: Paper (arXiv).
  5. E1

    RT-1 lab kitchen (EDR mobile manip) ≠ IFR factory stock / humanoid demos. RT-1 as baseline ≈0 on ALOHA fine tasks (Zhao et al.) is a different setup — do not collapse protocols.

    Brohan et al. RT-1. RT-1: Robotics Transformer for Real-World Control at Scale. https://arxiv.org/pdf/2212.06817. As of 2022-12. Checked 2026-09-09. Type: Paper (arXiv).
  6. E1

    RT-2: no new robotic motions from web pretrain; weak at grasp-by-part (e.g. handle), novel tool use / towel wipe, towel folding, multi-layer reasoning (Appendix G / §5). Lab ≠ factory.

    Brohan et al. RT-2. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. https://arxiv.org/pdf/2307.15818. As of 2023-07. Checked 2026-09-10. Type: Paper (arXiv).
  7. E1

    π0 lab/demo setting ≠ unsupervised factory AGI; vendor Physical Intelligence; open checkpoints noted in the ecosystem, not re-verified here; π0 ≠ warrant for general household autonomy.

    Black et al. π0. π₀: A Vision-Language-Action Flow Model for General Robot Control. https://arxiv.org/pdf/2410.24164. As of arXiv 2024-10 (v4 Jan 2026). Checked 2026-09-12. Type: Paper (arXiv).
  8. E1

    Thread Velcro 20% final success shows a vision/precision limit under the paper protocol — even with high success on Slot Battery (96%).

    Zhao et al. ALOHA/ACT. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. https://arxiv.org/pdf/2304.13705. As of 2023-04. Checked 2026-09-08. Type: Paper (arXiv).
  9. E0

    Tesla Optimus: tesla.com/AI access denied — no warrant in this entry. Figure BotQ 12k/year not opened (E0).

6. Actors and incentives

Who counts, who demos

  1. E1

    IFR counts industrial and service units with the ISO 8373 definition and a 12-year stock.

    IFR WR 2025 Exec. World Robotics 2025 — Industrial Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Industrial_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.IFR Sources. World Robotics 2025 — Sources and Methods. https://ifr.org/img/worldrobotics/Sources___Methods_WR_2025_Industrial_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.IFR Service Exec. World Robotics 2025 — Service Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Service_Robots.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.
  2. E3

    Figure and Boston Dynamics are demo sources; their caveats sit on the same pages.

    Figure F.02. F.02 Contributed to the Production of 30,000 Cars at BMW. https://www.figure.ai/news/production-at-bmw. As of 2025-11-19. Checked 2026-08-28. Type: Company report.Atlas CES. Boston Dynamics Unveils New Atlas Robot to Revolutionize Industry. https://bostondynamics.com/blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/. As of 2026-01-05. Checked 2026-08-28. Type: Company report.

7. State of the dispute

Humanoid hype vs. IFR stock

  1. E2

    Gu survey 2025 maps locomotion/manipulation challenges. IFR: no massive use. Survey ≠ deployment statistics.

    IFR SR slides. World Robotics 2025 — Service Robots Presentation. https://ifr.org/downloads/press_docs/Press_Conference_2025_SR.pdf. As of 2025. Checked 2026-08-28. Type: Association PDF.IFR humanoid info. Humanoid Robots — Vision and Reality (Infographic). https://ifr.org/downloads/press_docs/Humanoids_Position_Infograph_2025.pdf. As of 2025-08. Checked 2026-08-28. Type: Association PDF.Gu et al. Humanoid Locomotion and Manipulation: Current Progress and Challenges in Control, Planning, and Learning. https://arxiv.org/html/2501.02116. As of 2025-01. Checked 2026-08-28. Type: Paper (survey).

8. Open questions

  1. IFR humanoid full PDF (registration).
  2. World Robotics yearbook full text.
  3. Tesla Optimus primary page (access denied).
  4. Figure BotQ 12,000/year — not opened.
  5. Billard/Kragic dexterity full text.
  6. π0 opened 2026-09-12 (source 25, E1 architecture / E2 qualitative out-of-box); open checkpoints not re-verified.
  7. Atlas spec PDF.
  8. Agility, Apptronik, 1X, Unitree primaries.

9. Changes

  • v1.5-draft2026-09-12: π0 (Black et al., arXiv:2410.24164, source 25, E1 architecture / E2 out-of-box quality) opened — PaliGemma + flow-matching action expert; >~10,000 h; 7 configurations / 68 tasks; laundry/bussing/box; lab ≠ factory/household AGI.
  • v1.4-draft2026-09-10: RT-2 (Brohan et al., arXiv:2307.15818, source 24, E1) opened — Table 4 unseen avg 62 vs RT-1 32; Table 5 emergent 60 vs 17; no new motions from web.
  • v1.3-draft2026-09-09: RT-1 (Brohan et al., arXiv:2212.06817, source 23, E1) opened — Table 2 seen 97 / unseen 76; EDR kitchen ≠ factory; protocol ≠ ALOHA.
  • v1.2-draft2026-09-08: ALOHA/ACT (Zhao et al., arXiv:2304.13705, source 22, E1) opened — Table I/II final success; lab ≠ factory; Thread Velcro 20%.
  • v1.1-draft2026-09-06: OpenVLA (Kim et al., source 20, E1) and Open X-Embodiment/RT-X (source 21, E1) opened — lab VLA/X-embodiment alongside IFR stock.
  • v1.0-draftFirst version from the 2026-08-28 verification log.

10. Sources

No. Source As of Checked Grade
1 International Federation of Robotics. World Robotics 2025 — Industrial Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Industrial_Robots.pdf. Type: Association PDF. E1
2 International Federation of Robotics. Global Robot Demand in Factories Doubled in 10 Years. https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years. Type: Association press release. E1
3 International Federation of Robotics. World Robotics 2025 — Sources and Methods. https://ifr.org/img/worldrobotics/Sources___Methods_WR_2025_Industrial_Robots.pdf. Type: Association PDF. E1
4 International Federation of Robotics. Global Robot Density in Factories Doubled in Seven Years. https://ifr.org/ifr-press-releases/news/global-robot-density-in-factories-doubled-in-seven-years. Type: Association press release. E1
5 International Federation of Robotics. World Robotics 2025 — Service Robots Executive Summary. https://ifr.org/img/worldrobotics/Executive_Summary_WR_2025_Service_Robots.pdf. Type: Association PDF. E1
6 International Federation of Robotics. Service Robot Sales Rebound. https://ifr.org/ifr-press-releases/news/service-robots-see-global-growth-boom. Type: Association press release. E1
7 International Federation of Robotics. World Robotics 2025 — Service Robots Presentation. https://ifr.org/downloads/press_docs/Press_Conference_2025_SR.pdf. Type: Association PDF. E2
8 International Federation of Robotics. Humanoid Robots: Vision and Reality Paper Published by IFR. https://ifr.org/ifr-press-releases/news/humanoid-robots-vision-and-reality-paper-published-by-ifr. Type: Association press release. E2
9 International Federation of Robotics. Humanoid Robots — Vision and Reality (Infographic). https://ifr.org/downloads/press_docs/Humanoids_Position_Infograph_2025.pdf. Type: Association PDF. E2
10 OpenAI; Akkaya, I.; Andrychowicz, M.; Chociej, M.; et al.. Solving Rubik's Cube with a Robot Hand. https://arxiv.org/html/1910.07113. Type: Paper. E1
11 Handa, A.; Allshire, A.; Makoviychuk, V.; Petrenko, A.; et al.. DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality. https://arxiv.org/html/2210.13702v2. Type: Paper. E1
12 Gu; Li; Shen; Yu; Xie; McCrory; Cheng; Shamsah; Griffin; Liu; Kheddar; Peng; Zhu; Shi; Nguyen; Cheng; Gao; Zhao. Humanoid Locomotion and Manipulation: Current Progress and Challenges in Control, Planning, and Learning. https://arxiv.org/html/2501.02116. Type: Paper (survey). E2
13 Aljalbout; Xing; Romero; Akinola; Garrett; Heiden; Gupta; Hermans; Narang; Fox; Scaramuzza; Ramos. The Reality Gap in Robotics: Challenges, Solutions, and Best Practices. https://arxiv.org/html/2510.20808. Type: Paper (survey). E1
14 Li; Wang; Xu; Ye; Chen. The Developments and Challenges towards Dexterous and Embodied Robotic Manipulation: A Survey. https://arxiv.org/html/2507.11840v2. Type: Paper (survey). E2
15 Figure AI. F.02 Contributed to the Production of 30,000 Cars at BMW. https://www.figure.ai/news/production-at-bmw. Type: Company report. E3
16 Figure AI. F.03 Arrives at BMW. https://www.figure.ai/news/f-03-at-bmw. Type: Company report. E3
17 Figure AI. Introducing Helix 02: Full-Body Autonomy. https://www.figure.ai/news/helix-02. Type: Company report. E3
18 Boston Dynamics. Boston Dynamics Unveils New Atlas Robot to Revolutionize Industry. https://bostondynamics.com/blog/boston-dynamics-unveils-new-atlas-robot-to-revolutionize-industry/. Type: Company report. E3
19 Boston Dynamics. Introducing Electric Atlas. https://bostondynamics.com/news/introducing-electric-atlas/. Type: Company report. E3
20 Kim, M. J.; Pertsch, K.; Karamcheti, S.; Xiao, T.; Balakrishna, A.; Nair, S.; Rafailov, R.; Foster, E.; Lam, G.; Sanketi, P.; Vuong, Q.; Kollar, T.; Burchfiel, B.; Tedrake, R.; Sadigh, D.; Levine, S.; Liang, P.; Finn, C.. OpenVLA: An Open-Source Vision-Language-Action Model. https://arxiv.org/pdf/2406.09246. Type: Paper (arXiv). E1
21 Open X-Embodiment Collaboration. Open X-Embodiment: Robotic Learning Datasets and RT-X Models. https://arxiv.org/pdf/2310.08864. Type: Paper (arXiv). E1
22 Zhao, T. Z.; Kumar, V.; Levine, S.; Finn, C.. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware. https://arxiv.org/pdf/2304.13705. Type: Paper (arXiv). E1
23 Brohan, A.; Brown, N.; Carbajal, J.; Chebotar, Y.; et al. (Robotics at Google / Everyday Robots / Brain). RT-1: Robotics Transformer for Real-World Control at Scale. https://arxiv.org/pdf/2212.06817. Type: Paper (arXiv). E1
24 Brohan, A.; Brown, N.; Carbajal, J.; Chebotar, Y.; et al. (Google DeepMind). RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control. https://arxiv.org/pdf/2307.15818. Type: Paper (arXiv). E1
25 Black, K.; Brown, N.; Driess, D.; … Levine, S.; et al. (Physical Intelligence). π₀: A Vision-Language-Action Flow Model for General Robot Control. https://arxiv.org/pdf/2410.24164. Type: Paper (arXiv). E1

11. Uncertainty log

Overall uncertainty of this entry, bound to the verification log of 2026-08-28 plus OpenVLA/OXE 2026-09-06, ALOHA/ACT 2026-09-08, RT-1 2026-09-09, RT-2 2026-09-10 and π0 2026-09-12 (source 25). 25 openings. Lab VLA/ACT/π0 ≠ factory. Not used as warrant: The Robot Report, Wikipedia Optimus, Tesla (access denied).

  • Established (layer 1): 542,076 installations 2024; stock 4.66 million; China 54%; ISO definition; Rubik 20%; DeXtreme ~28; ALOHA/ACT Table I/II; RT-1 Table 2 (EDR); RT-2 Table 4/5 VLA (unseen 62; emergent 60); OXE 22 embodiments; OpenVLA +16.5 pp vs RT-2-X (lab); π0 architecture PaliGemma+flow matching, >~10,000 h, 7/68 (E1).
  • Claimed (layer 2/3): IFR 575k/700k; Figure F.02/Helix; Atlas CES — vendor.
  • Constrained (layer 3): humanoids vs. industry on speed; battery ≠ workday; demo ≠ rate; VLA lab/π0 ≠ factory/household AGI; Tesla E0.

Not opened (not a warrant)

  • IFR humanoid full PDF (registration).
  • World Robotics yearbook.
  • tesla.com/AI — access denied.
  • Figure BotQ 12k/year E0.
  • Billard/Kragic full text.
  • π0 opened 2026-09-12 (source 25); open checkpoints not re-verified; OXE/OpenVLA/ALOHA/ACT/RT-1/RT-2 opened earlier.
  • Atlas spec PDF.
  • Agility, Apptronik, 1X, Unitree.