Physical AI

Research

Ten cross-cutting docs synthesizing the state of Physical AI in 2026 — AV industry, robotics foundation models, simulation, labeling, world models, open problems, and a sequenced learning roadmap.

  1. 00

    Overview — Physical AI in 2026

    ~30 min

    Cross-cutting synthesis of the field: data is the bottleneck, the data engine is the product, the modular/E2E pendulum is dissolving.

  2. 01

    AV industry and data

    ~90 min

    Who collects what at what scale: Tesla, Waymo, Mobileye, Wayve, Waabi. Fleet logs vs customer-shadow vs sim vs world models.

  3. 02

    Robotics foundation models

    ~60 min

    The VLA recipe — RT-X, OpenVLA, π0/π0.5, Helix, Gemini Robotics. Open X-Embodiment, demo collection economics.

  4. 03

    Simulation and synthetic data

    ~90 min

    The competitor map. Applied Intuition Simian, Nvidia Cosmos + Isaac, CARLA, Waymax, the death of pure-rendering shops.

  5. 04

    Labeling and data curation

    ~120 min

    The most important doc for the role. Data-engine philosophy, FiftyOne/SAM2/OpenSCENARIO tools, foundation models as labelers.

  6. 05

    World models and generative

    ~60 min

    Cosmos, GAIA, GR00T-Dreams, Wayve PRISM-1. World models as data engine and as eval substrate.

  7. 06

    Open problems and benchmarks

    ~60 min

    The RAND 11-billion-mile result; SOTIF, ODD, V&V; benchmarks that drive the field forward.

  8. 07

    Learning roadmap with mini-projects

    ~45 min

    The 8-to-12-week plan. Each phase produces a tangible artifact; each project has a definite "done" criterion.

  9. 08

    Connections and gaps

    ~30 min

    How the projects link together — the critical chain, the round-2 audit, the bijective folder-rename mapping.

  10. 09

    Research frontier and outlook

    ~50 min

    What's next: open research questions, where the next 12–24 months of progress will land, what to watch.