PhysicalAI.bestOur standard
PHYSICAL AI, IN FOCUS

AI is leaving
the screen.
Track where it lands.

Robots, models, benchmarks, deployments and infrastructure — connected by evidence.

Start withHumanoids VLA Warehouse LIBERO
Explore the database Compare systems

Source-linked facts. No guessed specs.

A CONNECTED VIEW OF THE FIELDLIVE REGISTRY
EVIDENCE GRAPH8 CONNECTED RECORDS
FOLLOW A CONNECTIONFigure AI
→ serves Home Assistance
Evidence · reviewed 2026-09-28

Stated target application; does not establish a paid deployment.

co-figure-ai ↗ co-figure-ai-2 ↗
FROM CLAIM TO CONTEXT

A product is only part of the story.

Who builds itWhat powers itWhere it works
Follow the evidence
67Companies
44Robots & systems
31Models
18Benchmarks
8Customer-confirmed deployments
254Sources
Counts include reviewed records that meet our publication standard. See the standard ↗
Explore the landscape

The whole stack. In one place.

Move from the machine to the model — and the evidence behind both.

Explore infrastructure

Humanoids

Walking, reaching, manipulating

Industrial & warehouse

Systems built around the task

Robot foundation models

Policies across embodiments

Vision-language-action

From observations to actions

World models

Predicting physical environments

Simulation & digital twins

Train, test, then validate

Edge compute

Intelligence on the machine

Autonomous systems

Movement beyond the factory

Beyond the demo

What have customers confirmed?

Dated accounts of specific tasks and deployment stages. A recent source review does not establish continued operation.

Inspect deployments
↳
Agility Robotics

GXO / SPANX Digit tote handling

Transfer totes from cobots to conveyors

Deployed revision: Digit hardware generation not disclosed; do not map to Digit 5

Customer confirmed

Paid deployment

Stage at the evidence date; continued operation requires separate confirmation.

Activity evidence2024-06-27Record reviewed
Sources 1
GXO multi-year agreement with Agility Robotics

GXO · Published 2024-06-27 · Source accessed 2026-09-28

Customer describes commercial live-warehouse operation after a pilot. Does not identify current Digit 5 hardware.

↳
Diligent Robotics

Cedars-Sinai Moxi hospital logistics

Move linens, retrieve samples and medication, transport belongings

Deployed revision: Moxi hardware revision not disclosed

Customer confirmed

Recurring operation

Stage at the evidence date; continued operation requires separate confirmation.

Activity evidence2026-03-18Record reviewed
Sources 1
Cedars-Sinai on robots supporting nurses

Cedars-Sinai · Published 2026-03-18 · Source accessed 2026-09-28

Hospital newsroom summarizes an interview with its own nursing leaders and explicitly confirms local use and count.

↳
Starship Technologies

Co-op / Starship Leeds grocery delivery

Local grocery delivery from Co-op stores

Deployed revision: Hardware generation not identified in customer announcement

Customer confirmed

Recurring operation

Stage at the evidence date; continued operation requires separate confirmation.

Activity evidence2023-07-20Record reviewed
Sources 1
Co-op and Starship expand delivery across Leeds

Co-op · Published 2023-07-20 · Source accessed 2026-09-28

A better shortlist

Compare the decision. Not the hype.

Curated matchups with exact versions, unknowns and sources attached to every row.

All comparisons
Humanoid manipulation
FI
vsAP

Figure 03 vs Apptronik Apollo

Both candidates combine humanoid hardware with manipulation intelligence. Their current positioning differs: household-oriented Figure 03 and industrial Apollo 2 should be evaluated against a specified task, not assumed interchangeable.

15 dimensions · cited row by row
Industrial material movement
DI
vsWA

Agility Digit vs UBTech Walker S2

These humanoid programs address industrial handling, but the current Digit record describes Digit 5 in development. The comparison separates current product specifications from dated operating evidence for earlier generations.

15 dimensions · cited row by row
Humanoid development platforms
UN
vsFO

Unitree G1 vs Fourier GR-2

G1 and GR-2 are relevant to teams evaluating humanoid platforms for development. Configuration, supported interfaces and available assistance matter more than a spec-sheet maximum in isolation.

15 dimensions · cited row by row
Benchmark intelligence

Every score needs a frame of reference.

Read the revision, task and test environment before drawing a conclusion.

Explore benchmarks
Context before scores. Results from different benchmarks are not directly comparable. A simulation result does not establish real-world reliability, safety or commercial availability.
Model / revisionBenchmark / metricReported resultEnvironmentEvidence & scope
OpenVLA-OFTOpenVLA-OFT, arXiv:2502.19645v1 Table I final row; wrist camera + proprioceptionLIBEROTask success rate (%) · LIBERO-Spatial97.6Reported 2025-02-27simulationSimulated Franka Emika PandaPrimary releaseVerified 2026-09-28
Sources 1
Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success, v1

Moo Jin Kim, Chelsea Finn and Percy Liang · Published 2025-02-27 · Source accessed 2026-09-28

Table I final row and evaluation protocol inspected. Factual per-suite results only; no leaderboard copied.

Methodology & comparability

Per-suite fine-tuning; best checkpoint selected from periodic evaluations; 500 trials per suite. Author-reported simulation, not an independent reproduction or physical reliability estimate.

Comparison group: oft-2502.19645v1-table1-wrist-proprio

Paper explicitly CC BY 4.0. Attributed factual result only; no dataset or leaderboard redistributed.

Verified 2026-09-28

OpenVLA-OFTOpenVLA-OFT, arXiv:2502.19645v1 Table I final row; wrist camera + proprioceptionLIBEROTask success rate (%) · LIBERO-Object98.4Reported 2025-02-27simulationSimulated Franka Emika PandaPrimary releaseVerified 2026-09-28
Sources 1
Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success, v1

Moo Jin Kim, Chelsea Finn and Percy Liang · Published 2025-02-27 · Source accessed 2026-09-28

Table I final row and evaluation protocol inspected. Factual per-suite results only; no leaderboard copied.

Methodology & comparability

Per-suite fine-tuning; best checkpoint selected from periodic evaluations; 500 trials per suite. Author-reported simulation, not an independent reproduction or physical reliability estimate.

Comparison group: oft-2502.19645v1-table1-wrist-proprio

Paper explicitly CC BY 4.0. Attributed factual result only; no dataset or leaderboard redistributed.

Verified 2026-09-28

OpenVLA-OFTOpenVLA-OFT, arXiv:2502.19645v1 Table I final row; wrist camera + proprioceptionLIBEROTask success rate (%) · LIBERO-Goal97.9Reported 2025-02-27simulationSimulated Franka Emika PandaPrimary releaseVerified 2026-09-28
Sources 1
Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success, v1

Moo Jin Kim, Chelsea Finn and Percy Liang · Published 2025-02-27 · Source accessed 2026-09-28

Table I final row and evaluation protocol inspected. Factual per-suite results only; no leaderboard copied.

Methodology & comparability

Per-suite fine-tuning; best checkpoint selected from periodic evaluations; 500 trials per suite. Author-reported simulation, not an independent reproduction or physical reliability estimate.

Comparison group: oft-2502.19645v1-table1-wrist-proprio

Paper explicitly CC BY 4.0. Attributed factual result only; no dataset or leaderboard redistributed.

Verified 2026-09-28

OpenVLA-OFTOpenVLA-OFT, arXiv:2502.19645v1 Table I final row; wrist camera + proprioceptionLIBEROTask success rate (%) · LIBERO-Long94.5Reported 2025-02-27simulationSimulated Franka Emika PandaPrimary releaseVerified 2026-09-28
Sources 1
Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success, v1

Moo Jin Kim, Chelsea Finn and Percy Liang · Published 2025-02-27 · Source accessed 2026-09-28

Table I final row and evaluation protocol inspected. Factual per-suite results only; no leaderboard copied.

Methodology & comparability

Per-suite fine-tuning; best checkpoint selected from periodic evaluations; 500 trials per suite. Author-reported simulation, not an independent reproduction or physical reliability estimate.

Comparison group: oft-2502.19645v1-table1-wrist-proprio

Paper explicitly CC BY 4.0. Attributed factual result only; no dataset or leaderboard redistributed.

Verified 2026-09-28

Our operating principle

Evidence
before volume.

A polished demo is not a deployment. A company claim is not independent validation. Unknown is an honest answer.

How we assess evidence
Primary spec

A specification published by the manufacturer.

Customer confirmed

The named customer confirms the reported activity.

Independent report

Reporting from an organization outside the vendor.

Company claim

A statement made by the company, kept in context.

Edited demo

A demonstration, without an operating history.

Not publicly disclosed

The public record does not support a value.

Connect the dots

One company. A whole network.

Follow sourced relationships from organizations to products, models and field deployments.

Read the market map
EVIDENCE GRAPH8 CONNECTED RECORDS
FOLLOW A CONNECTIONFigure AI
→ serves Home Assistance
Evidence · reviewed 2026-09-28

Stated target application; does not establish a paid deployment.

co-figure-ai ↗ co-figure-ai-2 ↗
Start with the work

What do you need a system to do?

All buyer use cases
Warehouse PickingPalletizingTote HandlingTruck UnloadingManufacturing AssemblyAutomotive ManufacturingWeldingSanding & Surface Finishing
The verified change log

Changes and research, dated separately.

Market events retain their source context. Research additions record when evidence entered this database.

Read the research

Reported market changes

No subsequent market change has been verified yet. The initial research snapshot appears below.

Newly reviewed records

These dates describe our research work, not a new product release or deployment.

Reviewed
Research review / addition

RoboCasa365

Registry review: environment release and task horizon must accompany benchmark comparisons.

Sources 1
robocasa/robocasa official repository

robocasa · 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.

Reviewed
Research review / addition

GXO / SPANX Digit tote handling

Initial deployment review; named customer evidence and bounded maturity captured.

Sources 2
GXO multi-year agreement with Agility Robotics

GXO · Published 2024-06-27 · Source accessed 2026-09-28

Customer describes commercial live-warehouse operation after a pilot. Does not identify current Digit 5 hardware.

Digit 5 product and development disclaimer

Agility Robotics · Publication date not disclosed · Source accessed 2026-09-28

Reviewed
Research review / addition

Cedars-Sinai Moxi hospital logistics

Initial deployment review; named customer evidence and bounded maturity captured.

Sources 2
Cedars-Sinai on robots supporting nurses

Cedars-Sinai · Published 2026-03-18 · Source accessed 2026-09-28

Hospital newsroom summarizes an interview with its own nursing leaders and explicitly confirms local use and count.

Moxi hospital robot

Diligent Robotics · Publication date not disclosed · Source accessed 2026-09-28

Signal, delivered

The Physical AI Change Log

A weekly digest of important verified changes.
New models. Real deployments. Better context.

Get the Change Log

Editorially independent. Sponsorship always labeled.