Physical AI is AI that senses the real world and moves a machine in it. The clearest examples running at scale in October 2026 are driverless robotaxis (Waymo lists 15 US service areas), driverless trucks (Aurora hauls paid freight with no one aboard in Texas and the Sun Belt), and warehouse robot fleets (Amazon has deployed more than 1 million robots). Humanoid robots have a few paid deployments and many pilots, and surgical robots are AI-assisted tools controlled by a surgeon. Each example below says how far along it is, using the operator’s own figures.
For the definition and how physical AI differs from chatbots, start with what physical AI is.
Physical AI examples at a glance
| Example | What the AI does | Status (Oct 2026) | Key figure |
|---|---|---|---|
| Waymo robotaxis | Drives a car with no one in the driver’s seat | Paid service | 15 US service areas |
| Aurora trucks | Drives Class 8 trucks on highways | Paid driverless freight | Target of 200 driverless trucks by end of 2026 |
| Kodiak trucks | Drives frac-sand trucks off-road | Paid driverless work | 35 trucks, 40,000+ driverless hours |
| Gatik box trucks | Drives store-delivery trucks | Driverless, no one aboard | DFW, Phoenix, NW Arkansas |
| Amazon robots | Moves shelves, sorts, picks items | In production | 1 million+ robots deployed |
| Locus Robotics | Guides picking robots through aisles | In production | 8 billion+ picks |
| MiR pallet jack | Finds and lifts pallets by vision | Sold product | Trained on 1.2 million+ images |
| Agility Digit | Moves totes with a humanoid | Paid deployments | 65,000 production hours |
| da Vinci | Surgeon-controlled robot with AI assistance | Clinical use | 11,710 systems installed |
Sources for each row are linked in the sections below.
On the road: robotaxis and trucks
Waymo’s service page lists 15 US areas where anyone can hail a car with no driver; our Waymo fares guide covers how pricing works. Baidu’s Apollo Go runs fully driverless paid rides in Chinese cities and in Dubai, per its second-quarter 2026 results.
Freight is further along than many people expect. Aurora’s second-quarter 2026 shareholder letter describes paid driverless runs on lanes such as Dallas–Laredo and Fort Worth–Phoenix and a goal of 200 driverless trucks by year-end. Kodiak’s customer Atlas Energy owned 35 driverless trucks hauling frac sand in the Permian Basin at the end of June 2026, with more than 40,000 hours of paid driverless operation, per Kodiak’s second-quarter results. Gatik runs box trucks with nobody aboard between warehouses and stores; see our Gatik guide.
In warehouses: robot fleets
Amazon said in July 2025 that it had deployed its one-millionth robot across 300+ facilities and that its DeepFleet AI model improves the fleet’s travel efficiency by 10%. Locus Robotics says its robots have made more than 8 billion picks, and Geek+ reported 81,000 robots deployed in its first-half 2026 results. These fleets are physical AI in a practical form: each robot plans its own path, and fleet software decides which robot does which job. Our guides to Amazon’s robots and AMR companies go deeper.
In factories: robots that see and adapt
Traditional factory robots repeat a taught path. The physical AI version adds perception so the robot copes with variation. MiR’s pallet jack finds and lifts pallets, including shrink-wrapped ones, using a detection model trained on more than 1.2 million real and synthetic images, per MiR’s launch article. Doosan’s PalletizHD+ plans palletizing paths with AI, and FANUC and other arm makers now link their robots to NVIDIA simulation tools for training. See our NVIDIA physical AI guide.
Humanoids: real work, mostly in pilots
Agility Robotics lists GXO, Schaeffler and Toyota Motor Manufacturing Canada as customers on its home page and says Digit has 65,000 hours of production experience moving totes. Most other humanoids are in pilots or demonstrations (our humanoid deployment tracker lists who is where); our humanoid robot companies page sorts makers by how far they’ve gotten, and the Figure AI guide covers one of the best-funded.
Homes are harder than factories. A robot can pick up a cup but still struggle to finish a chore, as we explain in why robots struggle to finish household chores and why feeding towels into a folding machine is hard.
In hospitals: surgeon-controlled robots
Intuitive’s da Vinci systems are controlled by a surgeon at a console, with software assisting rather than operating on its own. Intuitive reported 11,710 systems installed worldwide at June 30, 2026, in its second-quarter 2026 results.
What separates real physical AI from a demo
Ask three questions of any example: Is it running for paying customers, a pilot, or a video? Is a person driving, supervising remotely, or absent? And what number does the operator publish (rides, miles, hours, picks)? The examples above all publish a number. Video-based models that learn about the world are a big research area, but training on video alone doesn’t yet give a robot the action data it needs; see where video learning ends and robot action data begins.
Operator filings and official pages checked October 24, 2026.