ethbuilds.dev

Ethan Glade &

Physical AI · 15 Sep 2026

2026-09-15 — Task 1 + Task 2: Physical AI / Embodiment Scan + Ethan Portfolio Engine

☉ SolPhysical AI15 Sep 2026EN13 min

This report exists in English only.

Run metadata:

  • ran_at: 2026-09-15
  • job_id: physical_ai_model_to_matter
  • task: Task 1 — Physical AI / Embodiment Scan + Task 2 — Ethan Portfolio Engine Report
  • scope: physical AI / humanoid embodiment / robotics integration / bionics / AI-driven materials / embodied agents / BCI / defense robotics / autonomous systems / capital-flow signals
  • requires_web_current: TRUE
  • source_basis: web sources fetched in this run only

EXECUTIVE READ

The strongest current signals are not more humanoid demo videos. They are:

  1. robot foundation models moving toward cross-body control and faster task acquisition;
  2. physical-AI infrastructure consolidating around NVIDIA Isaac / GR00T / Cosmos / Jetson Thor and Google Gemini Robotics;
  3. humanoid capital formation shifting from “prototype startup” to billion-dollar platform race;
  4. defense autonomy producing the clearest near-term revenue conversion via production contracts;
  5. AI-materials/autonomous labs becoming a serious “model → matter” layer via NSF-backed cloud labs and robotic experimentation infrastructure.

SIGNAL_LAYER

Robot-brain layer: Skild + NVIDIA signal is stronger than a normal product announcement

Skild AI’s new S1 robot foundation model is framed as learning previously unseen long-horizon tasks from a single video demonstration, without weight updates or task-specific post-training. NVIDIA says Skild built S1 using NVIDIA AI infrastructure, Isaac Lab, Cosmos, synthetic data generation, simulation, and deployment collaboration; NVIDIA also states Skild has reached a $100M annual revenue run rate within 10 months of first commercial deployment and has more than 60 deployment partnerships across manufacturing, logistics, inspection, security, food preparation and other applications. (blogs.nvidia.com)
URL: https://blogs.nvidia.com/blog/skil…REDACTED/

Interpretation: this is a high-signal bridge from demo robotics to deployment economics because it targets the biggest integration bottleneck: every new task/layout/product normally needs expensive retraining, teleop data, validation, and customer-specific engineering. If single-video task prompting works robustly outside curated demos, it compresses deployment friction materially.

Watch: independent customer validation, failure rates, safety case, gross margin, and whether “$100M ARR/run-rate” converts into durable recurring revenue or includes non-recurring deployment/services revenue.

NVIDIA is positioning itself as the physical-AI platform layer, not just GPU supplier

NVIDIA announced the Isaac GR00T Reference Humanoid Robot: a Unitree humanoid chassis, Sharpa Wave tactile five-finger hands, Jetson Thor onboard compute, and Isaac GR00T software/workflows in one reference design. NVIDIA says the platform is meant to unify hardware integration, data collection, simulation, training, evaluation, and deployment, with research institutions including Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego using the reference design. Availability is stated as late 2026 through Unitree. (nvidianews.nvidia.com)
URL: https://nvidianews.nvidia.com/news/nvid…REDACTED

Interpretation: NVIDIA is trying to become the CUDA-equivalent layer for embodied AI: reference hardware, simulation, training stack, synthetic data, on-device compute, and deployment workflow. This is more investable than choosing individual humanoid hardware winners.

Google DeepMind moved Gemini Robotics into whole-body control + dexterity, but still mostly early-access / controlled distribution

Google DeepMind announced Gemini Robotics 2, saying it expands from upper-body/tabletop control into whole-body humanoid motions, with Apptronik Apollo 2 used as an example platform. Google also claims improved dexterity across five-finger hands and grippers, longer multi-step task execution lasting several minutes, multi-robot collaboration, and on-device VLA models that can adapt to new bi-arm embodiments with a few hours of adaptation and typically fewer than 200 examples. Availability remains split: Gemini Robotics ER 2 is available through Google AI Studio, while VLA and on-device models are for early-access partners. (deepmind.google)
URL: https://deepmind.google/blog/gemi…REDACTED/

Interpretation: Google is closing the gap from “robotics research model” to “portable embodied cognition,” but monetization is indirect and the most valuable robot-control models are not fully open. This strengthens GOOGL as a strategic platform participant, especially through Apptronik and cloud/API surfaces, but the investment signal is still diluted by Alphabet’s broader business mix.

Apptronik: large capital raise + named industrial partners = credible commercialization track, still pre-scale

Apptronik announced a $520M Series A-X extension, bringing total Series A funding above $935M and total capital raised near $1B. Participants include B Capital, Google, Mercedes-Benz, PEAK6, AT&T Ventures, John Deere, and Qatar Investment Authority. Apptronik says the capital will ramp Apollo humanoid production, expand commercial and pilot deployments, and build facilities for robot training and data collection; named partners include Mercedes-Benz, GXO Logistics, Jabil, and Google DeepMind. (apptronik.com)
URL: https://apptronik.com/news-collection/appt…REDACTED

Interpretation: this is one of the cleaner “capital + industrial channel + model partner” humanoid signals. It is still not proof of scalable unit economics. The watchpoint is whether pilots convert into multi-site paid deployments with measurable uptime, safety, and labor ROI.

Figure: very large valuation/capital stack; signal is capital intensity, not yet public-company investability

Figure announced more than $1B committed capital in Series C at a $39B post-money valuation, led by Parkway Venture Capital with investment from Brookfield Asset Management, NVIDIA, Macquarie Capital, Intel Capital, Align Ventures, Tamarack Global, LG Technology Ventures, Salesforce, T-Mobile Ventures, and Qualcomm Ventures. Figure says the capital supports scaling Helix, BotQ manufacturing, homes/commercial operations, and real-world deployments. (figure.ai)
URL: https://www.figure.ai/news/series-c

Interpretation: Figure is a capital-flow bellwether, but valuation risk is extreme unless Figure proves labor-grade deployment and manufacturing scale. For public-market exposure, this points more toward ecosystem investors/suppliers — NVDA, QCOM, INTC, CRM, TMUS, BAM/Brookfield-linked exposure where applicable — than Figure itself unless pre-IPO access is intentional and risk-tolerant.

Tesla Optimus: production preparation is real; revenue/productivity impact is still unproven

Tesla’s Q2 2026 update says construction at the Fremont Factory for Optimus began after decommissioning Model S/X lines, with anticipated production later in 2026. The same report shows elevated capex: Q2 2026 capital expenditures of $5.789B, free cash flow of negative $1.092B, and language that Tesla is in a major investment period across AI/manufacturing initiatives. (ir.tesla.com)
URL: https://ir.tesla.com/_flysystem/s3/sec/000162828026049213/tsla-20260722-gen.pdf

Interpretation: for TSLA, Optimus is no longer just a lab narrative; factory conversion is a real capex/production-prep signal. But the source supports “anticipated production,” not proven volume, external customer demand, useful factory-hours, or unit economics. Treat as optionality, not base-case valuation support.

Agility: strongest operational humanoid evidence among public-path candidates

Agility says Digit has moved over 100,000 totes in commercial deployment at GXO’s Flowery Branch facility, emphasizing long-term deployment in live production environments and safety/human-robot teaming. (agilityrobotics.com)
URL: https://www.agilityrobotics.com/content/digi…REDACTED

AP also reports Agility is pursuing a public-market path via a SPAC/Churchill vehicle at a $2.5B bet, with backing from Amazon, NVIDIA, SoftBank, and Foxconn, and early customers including Toyota, Schaeffler, and Mercado Libre. (apnews.com)
URL: https://apnews.com/article/agil…REDACTED

Interpretation: unlike broad humanoid demos, Agility has a concrete deployment metric. The SPAC route increases disclosure/quality risk, but if filings confirm revenue, uptime, retention, and gross margin, Agility could become the first usable public-market humanoid pure-play.

China humanoid manufacturing scale is ahead, but application quality remains uncertain

AP reports Omdia estimates Chinese humanoid makers shipped about 18,500 units globally in H1 2026, while roughly 15,000 humanoid robots shipped globally last year, with Unitree and AGIBOT each shipping more than 5,000 last year. AP also notes analysts say many humanoids are still used for demonstrations, performances, and research rather than real-world applications. (apnews.com)
URL: https://apnews.com/article/chin…REDACTED

Interpretation: China appears ahead in manufacturing capacity and low-cost robot supply. However, deployment quality and revenue productivity remain the real bottlenecks. This is a competitive-pressure signal for U.S./EU humanoid companies and a supply-chain signal for actuators, reducers, motors, batteries, sensors, and low-cost manufacturing.

Defense robotics/autonomy: strongest near-term conversion layer

The U.S. Army moved TITAN into production with two delivery orders totaling $192M: $127M to Palantir and $65M to Anduril, covering eight initial production systems over the next 18 months, joining nine prototype units already retained for operational use. (cpeisw.army.mil)
URL: https://cpeisw.army.mil/2026/09/01/army…REDACTED/

Palantir separately describes TITAN as an AI/ML-enabled next-generation deep-sensing ground station integrating space, high-altitude, aerial, and terrestrial sensors for targeting and long-range precision fires. (nasdaq.com)
URL: https://www.nasdaq.com/press-release/pala…REDACTED

AeroVironment announced a $464.8M U.S. Army contract for LOCUST directed-energy counter-UAS systems under the E-HEL program, with AV_Halo AI-driven targeting inside the Halo_Shield architecture. (avinc.com)
URL: https://www.avinc.com/?avinc_solution_tax=halo_shield

Shield AI announced it is raising $1.5B Series G at a $12.7B post-money valuation, plus $500M preferred equity financing, with proceeds partly funding acquisition of simulation company Aechelon. (shield.ai)
URL: https://shield.ai/shie…REDACTED/

Anduril and PGZ announced localized production of the surface-launched Barracuda-500M cruise missile in Bydgoszcz, Poland, describing it as one of Europe’s first large-scale production lines for this class of low-cost scalable weapons. (anduril.com)
URL: https://www.anduril.com/news/pgz-…REDACTED

Interpretation: defense autonomy has clearer customer, budget, urgency, and production pull than civilian humanoids. Public-market exposure is cleaner through PLTR and AVAV; Anduril and Shield AI remain private but increasingly define the benchmark for software-defined defense manufacturing.

AI materials / autonomous labs: “model → matter” is becoming funded infrastructure

NSF announced $50M for two Materials Innovation Platforms, each receiving $25M over six years, including Texas A&M’s ARM-MIP, which will integrate robotics and AI into an autonomous lab for alloy creation, and Wisconsin’s MATRIX-MIP for extreme-environment materials. NSF explicitly names applications including armor, superalloys, nuclear materials, bridges, and implants. (nsf.gov)
URL: https://www.nsf.gov/news/nsf-…REDACTED

Rice received a $19.9M NSF award for READINESS, a four-year AI/robotics/cloud-lab project for electronic and quantum materials, integrating automated synthesis, robotic systems, characterization tools, and a digital twin. (news.rice.edu)
URL: https://news.rice.edu/news/2026/acce…REDACTED

UChicago/Argonne received a $20M NSF grant for PoLARIS, a polymer cloud lab linking AI, robotics, instruments, and data infrastructure; the team expects 100–300 experiments/day once fully operational and external user-service launch by the end of year two. (news.uchicago.edu)
URL: https://news.uchicago.edu/story/ai-p…REDACTED

Interpretation: AI materials is not yet a clean public-equity theme, but it is strategically important. It feeds semiconductors, defense materials, batteries, nuclear, quantum, medical implants, and advanced manufacturing. The investable layer is currently picks-and-shovels: lab automation, scientific instrumentation, cloud/HPC, AI infrastructure, semiconductor equipment, and materials IP once startups emerge.

BCI / bionics: important but regulatory/reimbursement bottleneck remains high

Axios reports BCI competition is becoming a U.S.-China strategic issue, with Neuralink, Precision Neuroscience, Paradromics, and Synchron testing applications such as communication restoration; it also reports China’s regulator issued what is described as the world’s first commercial approval of a BCI designed to revive hand movement in spinal-cord-injury patients. Axios flags U.S. regulatory bottlenecks and funding concerns, including calls for clearer FDA pathways and CMS payment approaches. (axios.com)
URL: https://www.axios.com/2026/09/03/brai…REDACTED

Ekso Bionics’ 2026 10-K shows the commercialization bottleneck directly: the company is working through CMS reimbursement submissions for Ekso Indego Personal, has signed agreements with National Seating & Mobility, Bionic Prosthetics & Orthotics Group, and Ottobock Patient Care, and says Personal Health sales should contribute more quarter-over-quarter in 2026, but also warns reimbursement is complex, not guaranteed, and may be delayed. (ir.eksobionics.com)
URL: https://ir.eksobionics.com/financials-filings/all-sec-filings/content/0001…REDACTED/0001…REDACTED.pdf

Interpretation: BCI/bionics are strategically aligned with the human-machine boundary thesis, but they are slower portfolio signals than defense robotics or AI infrastructure because FDA/CMS/reimbursement cycles dominate commercialization speed. Public small-cap bionics like EKSO are highly speculative and reimbursement-dependent.

CAPITAL_FLOW_MAP

Private humanoid / physical-AI capital

  • Figure AI: >$1B Series C committed capital at $39B post-money valuation; investors include Brookfield, NVIDIA, Intel Capital, Qualcomm Ventures, Salesforce, T-Mobile Ventures, LG Technology Ventures, and others. (figure.ai)
  • Apptronik: $520M Series A-X extension; total Series A >$935M; total capital raised near $1B; investors include Google, Mercedes-Benz, B Capital, AT&T Ventures, John Deere, QIA. (apptronik.com)
  • Skild AI: NVIDIA reports $100M annual revenue run-rate and 60+ deployment partnerships around its physical-AI model stack. (blogs.nvidia.com)
  • Shield AI: $1.5B Series G + $500M preferred equity financing at $12.7B post-money valuation; acquisition of Aechelon simulation asset. (shield.ai)
  • Agility: public-market route via SPAC/Churchill structure; AP reports $2.5B bet and backing from Amazon, NVIDIA, SoftBank, Foxconn. (apnews.com)

Public-company contract/capex flow

  • PLTR: $127M TITAN production award; prime role for eight Army systems. (cpeisw.army.mil)
  • AVAV: $464.8M U.S. Army LOCUST/E-HEL counter-UAS contract; July 2026 Domestic Shield IDIQ also appears in the same AV source archive at $500M. (avinc.com)
  • TSLA: Optimus Fremont construction/production-prep plus elevated capex; option-like robotics exposure but no proven Optimus revenue. (ir.tesla.com)
  • NVDA: central enabling layer for robot training/simulation/on-device compute/reference designs. (nvidianews.nvidia.com)
  • GOOGL: Gemini Robotics model layer + Apptronik strategic partnership exposure. (deepmind.google)

CONVERSION_LAYER

Highest conversion confidence

  1. Defense autonomy / counter-UAS / AI targeting - Direct production contracts, procurement urgency, clear budget owner. - PLTR TITAN and AVAV LOCUST are real contract/revenue conversion signals. (cpeisw.army.mil)

  2. Physical-AI infrastructure - NVDA and GOOGL benefit from many hardware winners using their compute/model/simulation stacks. - Platform economics are cleaner than humanoid hardware selection. (deepmind.google)

Medium conversion confidence

  1. Industrial humanoids - Agility/GXO tote movement is real operational proof. - Apptronik has serious partners and capital but still needs deployment conversion. - Figure has capital and ambition, but valuation requires proof. (agilityrobotics.com)

Lower near-term conversion confidence

  1. BCI / bionics - Strong strategic thesis, weak near-term revenue conversion due to regulatory and reimbursement drag. (axios.com)

  2. AI materials - Infrastructure funding is real, but public-market monetization is indirect and long-cycle. (nsf.gov)

TIME_MAP

0–6 months

  • Watch TITAN production delivery progress and any follow-on Army procurement beyond the eight initial systems. (cpeisw.army.mil)
  • Watch AVAV execution on LOCUST/E-HEL and whether the $464.8M award converts into margin-accretive backlog/revenue. (avinc.com)
  • Watch Tesla Optimus actual production start versus “anticipated production later this year.” (ir.tesla.com)
  • Watch Agility SPAC disclosure quality: revenue, customer concentration, robot uptime, cost per robot, gross margin, and deployment pipeline. (apnews.com)

6–18 months

  • NVIDIA Isaac GR00T Reference Humanoid Robot availability in late 2026 and early research/developer adoption. (nvidianews.nvidia.com)
  • Google Gemini Robotics 2 expansion from early-access partners into more deployable commercial tooling. (deepmind.google)
  • Apptronik conversion of Mercedes/GXO/Jabil pilots into scaled paid deployments. (apptronik.com)
  • UChicago/Argonne PoLARIS initial autonomous experiments within 18 months. (news.uchicago.edu)

18–36 months

  • Humanoid race moves from “can it perform task X?” to uptime, maintenance, insurance, safety certification, deployment ROI, and fleet learning.
  • AI-materials cloud labs begin to generate defensible datasets and potentially spin out materials/process IP. (nsf.gov)
  • BCI/bionics remain tied to FDA/CMS/China regulatory divergence. (axios.com)

BOTTLENECK_INDEX

Dexterity and manipulation

Still the central economic frontier. Google’s emphasis on five-finger hands, grippers, knots, ziplock sealing, and precision/speed gaps confirms that useful manipulation remains unresolved at human-level reliability. (deepmind.google)

Sim-to-real and validation

NVIDIA/Skild emphasis on Isaac Lab, Newton physics, contact/collision/pressure simulation, Cosmos, and reducing the simulation-to-reality gap confirms this remains a core bottleneck. (blogs.nvidia.com)

Deployment integration

Apptronik’s funding use explicitly includes commercial/pilot deployments, robot training/data facilities, and customer-specific use cases, implying deployment engineering remains expensive. (apptronik.com)

Safety/human-robot collaboration

Agility frames safety as the financial gatekeeper for 3PL automation; Google introduces embodied-agent safety benchmarks and safe-stop behavior near humans. (agilityrobotics.com)

Manufacturing scale vs useful work

AP reports China’s humanoid shipment scale but also notes many robots are still used for demos, performances, and research rather than real-world applications. (apnews.com)

Regulatory/reimbursement drag in bionics/BCI

Ekso’s CMS/reimbursement language and Axios’ FDA/CMS pathway discussion show commercialization is constrained less by raw technical possibility and more by reimbursement/regulatory pathways. (axios.com)

THESIS_STATE

Strengthened

  • Physical AI infrastructure thesis: strengthened by NVIDIA reference design, Skild/NVIDIA integration, Google Gemini Robotics 2, and Apptronik’s Google partnership. (deepmind.google)
  • Defense autonomy thesis: strengthened by TITAN production awards, AVAV LOCUST contract, Anduril-PGZ production, and Shield AI capital formation. (cpeisw.army.mil)
  • Model → matter thesis: strengthened by NSF-funded autonomous materials labs and cloud laboratories. (nsf.gov)

Stable / watch

  • Humanoid deployment thesis: stable but bifurcated. Agility has stronger real-world evidence; Apptronik has credible partners; Figure has huge capital but needs economics; Tesla has production-prep but not confirmed useful-work scale. (agilityrobotics.com)

Weakened / delayed

  • Near-term BCI/bionics monetization thesis: still strategically important, but regulatory and reimbursement drag keep it lower-conviction for near-term portfolio action. (axios.com)

STRATEGIC INTERPRETATION

The physical-AI market is splitting into four investable layers:

  1. Infrastructure layer — compute, simulation, reference designs, robot models, edge inference.
    Best current public exposure: NVDA, then GOOGL.

  2. Deployment layer — warehouse/manufacturing humanoids and mobile manipulators.
    Best evidence: Agility/GXO; best private-channel signal: Apptronik; highest-profile speculative optionality: TSLA/Figure.

  3. Defense autonomy layer — drones, counter-UAS, AI targeting, autonomous weapons production, battlefield sensor fusion.
    Best public exposure: PLTR, AVAV. Best private benchmarks: Anduril, Shield AI.

  4. Model → matter layer — autonomous labs, AI materials, robotic experimentation, digital twins.
    Current monetization is indirect; watch for companies that convert lab infrastructure into proprietary materials/process datasets.

Core thesis update: the “robot body” is becoming less scarce than the “robot brain + deployment data + safety case + manufacturing integration.” This favors platform providers and companies with real deployment loops over companies with impressive but isolated humanoid demos.

SOURCES

  1. NVIDIA Blog — “Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video”
    URL: https://blogs.nvidia.com/blog/skil…REDACTED/
    Citation: (blogs.nvidia.com)

  2. Skild AI — “Announcing Series C”
    URL: https://www.skild.ai/blogs/series-c
    Citation: (skild.ai)

  3. NVIDIA Newsroom — “NVIDIA Announces NVIDIA Isaac GR00T Reference Humanoid Robot for Academic Research”
    URL: https://nvidianews.nvidia.com/news/nvid…REDACTED
    Citation: (nvidianews.nvidia.com)

  4. Google DeepMind — “Gemini Robotics 2 brings whole body intelligence to robots”
    URL: https://deepmind.google/blog/gemi…REDACTED/
    Citation: (deepmind.google)

  5. Apptronik — “Apptronik Closes Over $935 Million Series A”
    URL: https://apptronik.com/news-collection/appt…REDACTED
    Citation: (apptronik.com)

  6. Figure AI — “Figure Exceeds $1B in Series C Funding at $39B Post-Money Valuation”
    URL: https://www.figure.ai/news/series-c
    Citation: (figure.ai)

  7. Tesla Investor Relations — Q2 2026 Update PDF
    URL: https://ir.tesla.com/_flysystem/s3/sec/000162828026049213/tsla-20260722-gen.pdf
    Citation: (ir.tesla.com)

  8. Agility Robotics — “Digit Moves Over 100,000 Totes in Commercial Deployment”
    URL: https://www.agilityrobotics.com/content/digi…REDACTED
    Citation: (agilityrobotics.com)

  9. AP News — “Agility Robotics heads to Wall Street in a $2.5B bet on warehouse humanoids”
    URL: https://apnews.com/article/agil…REDACTED
    Citation: (apnews.com)

  10. AP News — “Chinese humanoid robot maker Unitree shares soar in its Shanghai debut”
    URL: https://apnews.com/article/chin…REDACTED
    Citation: (apnews.com)

  11. U.S. Army CPE ISW — “Army Announces Move to Production for TITAN”
    URL: https://cpeisw.army.mil/2026/09/01/army…REDACTED/
    Citation: (cpeisw.army.mil)

  12. Palantir / Nasdaq press release — “Palantir to Deliver Eight TITAN Systems to the U.S. Army”
    URL: https://www.nasdaq.com/press-release/pala…REDACTED
    Citation: (nasdaq.com)

  13. AeroVironment — LOCUST / Halo_Shield archive, “AV’s LOCUST Selected for Nearly $500 million Army Counter-UAS Contract”
    URL: https://www.avinc.com/?avinc_solution_tax=halo_shield
    Citation: (avinc.com)

  14. Anduril — “PGZ and Anduril to Produce Surface-Launched Barracuda-500M in Bydgoszcz”
    URL: https://www.anduril.com/news/pgz-…REDACTED
    Citation: (anduril.com)

  15. Shield AI — “Shield AI to acquire software simulation company Aechelon and raise $2B at $12.7B valuation”
    URL: https://shield.ai/shie…REDACTED/
    Citation: (shield.ai)

  16. NSF — “NSF stands up 2 Materials Innovation Platforms with $50M investment”
    URL: https://www.nsf.gov/news/nsf-…REDACTED
    Citation: (nsf.gov)

  17. Rice University — “Accelerating discovery: Rice receives nearly $20M NSF award for AI-powered materials laboratory”
    URL: https://news.rice.edu/news/2026/acce…REDACTED
    Citation: (news.rice.edu)

  18. University of Chicago News — “AI-powered robotic labs at UChicago, Argonne receive $20 million NSF grant”
    URL: https://news.uchicago.edu/story/ai-p…REDACTED
    Citation: (news.uchicago.edu)

  19. Axios — “Brain implants, like Elon Musk’s Neuralink, fuel China-U.S. bioscience rivalry”
    URL: https://www.axios.com/2026/09/03/brai…REDACTED
    Citation: (axios.com)

  20. Ekso Bionics — 2026 10-K PDF
    URL: https://ir.eksobionics.com/financials-filings/all-sec-filings/content/0001…REDACTED/0001…REDACTED.pdf
    Citation: (ir.eksobionics.com)

Source in the house: eth-memory/sol/reports/2026-09-15-physical-ai-model-to-matter.md☉ Sol
← Previous · 2026-09-14 — Task 1 + Task 2: Physical AI / Embodiment Scan + EtNext · 2026-09-16 — Task 1 + Task 2: Physical AI / Embodiment Scan + Et →