Map the market by the function a company or project supplies, then follow its products, models and evidence. This avoids mixing a research checkpoint, robot manufacturer, software platform and delivery operator into one misleading ranking.
Identity boundary. Original PhysicalAI.best editorial guide. Source links support factual examples; evaluation questions and decision rules are editorial guidance, not upstream endorsements.
Classify the offering before the company
Use the market map as a navigation aid. The relevant object may be a robot system, model, simulator, component, application platform or operated service. Companies can occupy several layers at once, so a single category should not erase the rest of their portfolio. For example, Vention’s design and controls offering addresses different decisions from a complete robot purchase or a delivery-service contract.
Waymo · Publication date not disclosed · Source accessed 2026-09-28
Official page read during live research; confirms identity and listed offering. Performance language remains a company claim.
Robot makers and installed systems
Hardware suppliers can sell a robot, an application cell or an integrated facility system. Boston Dynamics’ task-specific portfolio and AutoStore’s storage-system components illustrate the distinction. Compare suppliers against the workflow and installation scope the buyer needs. Do not rank a mobile inspection robot against a warehouse storage system merely because both are grouped under physical AI.
AutoStore · Publication date not disclosed · Source accessed 2026-09-28
Official page read during live research; confirms identity and listed offering. Performance language remains a company claim.
Models and autonomy software
Model developers and autonomy-platform suppliers occupy different positions from manufacturers. Google DeepMind documents action and reasoning models, while Brain Corp supplies software associated with partner hardware and commercial operations. A map should show who develops the intelligence, what hardware it is actually connected to and how access works. A claimed multi-embodiment model still needs evidence for each product integration.
Brain Corp · Publication date not disclosed · Source accessed 2026-09-28
Official page read during live research; confirms identity and listed offering. Performance language remains a company claim.
Tools, data and simulation
Research tools can be useful to developers without being turnkey systems for an industrial buyer. MuJoCo, LeRobot and dataset projects sit at different parts of the development workflow. Their public availability should be recorded alongside maintenance, integration and license conditions. Avoid treating open code, open weights, open datasets and a freely accessible paper as interchangeable forms of access.
google-deepmind · Publication date not disclosed · Source accessed 2026-09-28
Official README inspected live. Repository code terms do not automatically cover model weights, data, or third-party assets.
Operators and end-user workflows
An operated service is judged through coverage, workflow, support and operating evidence as well as the underlying robot. Waymo’s passenger service and Starship’s delivery partnerships illustrate this layer. Distinguish the operator, hardware developer and customer where the sources allow. A consumer using a service does not automatically have access to buy or license the system behind it.
Starship Technologies · Publication date not disclosed · Source accessed 2026-09-28
Official page read during live research; confirms identity and listed offering. Performance language remains a company claim.
Use commercial maturity as a separate dimension
Keep research release, demonstration, pilot, paid deployment and recurring operation separate from market category. GXO’s Digit commercial account is evidence about a defined workflow; current Digit 5 development remains a separate record. A company can simultaneously operate one product and develop another. Apply the maturity label to the particular system or deployment rather than to the entire organization.
Agility Robotics · Publication date not disclosed · Source accessed 2026-09-28
Read gaps as research priorities
An absent price, model version or deployment metric is information about current evidence coverage. Follow the graph to identify which source would resolve it: technical documentation, a customer statement, a dated quote or a reproducible evaluation. Use the map to build a task-specific shortlist and update queue. Company visibility, financing and sponsorship should not determine evidence quality or comparison outcomes.
Related reading is an editorial crosslink. Sourced connections describe relationships reported in the cited material. A link to a versioned profile does not establish compatibility with that version unless the connection note explicitly identifies it.
Record history & verified changes
A research review records when we checked a source. It does not mark a product launch or a new deployment.
Initial reviewed record. No subsequent field change has been recorded.
google-deepmind · Repository · Publication date not disclosed · Source accessed 2026-09-28
License: Apache-2.0 repository software; CC BY 4.0 other repository content; constituent dataset licenses separate. Factual summary and attribution only; no upstream prose, images, weights, or dataset redistributed.
Official README inspected live. Repository code terms do not automatically cover model weights, data, or third-party assets.