Physical AI & Robotics — AI Maturity

The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.

AI Maturity by Domain

Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail

DOMAIN

BLEEDING EDGEESTABLISHED

🦾 Physical AI & Robotics

AI controlling physical systems across manufacturing, agriculture, construction, healthcare, and service. The largest domain at 27 practices, with industrial robotics and pick-and-place at good-practice. Most practices cluster at leading-edge — sim-to-real transfer, humanoid robotics, and surgical automation are progressing rapidly. Nearly half the practices are advancing, making this one of the most dynamic domains by momentum.

27 practices: 8 good practice, 17 leading edge, 2 bleeding edge

Physical AI & Robotics — Biweekly Brief

The headline: Robots that do one narrow job in a controlled setting are now reliably profitable; general-purpose humanoids are attracting record orders but still failing in the real world. The money is in the boring tasks, not the ones that make headlines.

The Picture

Robotics has split into two camps, and where you sit determines your risk. Most companies running robots for a single, repetitive task in a predictable environment — welding, inspecting parts for defects, picking warehouse orders, spotting equipment that is about to fail — are getting real returns, often paying back the investment in well under a year. A smaller, louder group is betting on general-purpose humanoid robots; the orders are enormous (Hyundai just committed to more than 25,000) but the technology still stumbles outside the lab. The rest face a closing window: rivals in China and South Korea are pulling ahead on the factory floor, and roughly 80 percent of US manufacturing sites still run no automation at all. The lesson for any operator is the same — start narrow, in a setting you control, and treat humanoid hype as a 2028-plus question, not a 2026 purchase.

This Fortnight

Hyundai placed the largest humanoid robot order on record — more than 25,000 units for its car plants by 2028. German supplier Schaeffler and others signed similar multi-year deals, and analysts now estimate a humanoid can pay for itself in six months when kept busy. The economics are turning real for narrow, high-volume factory jobs — but only those; this is not yet a general-purpose workforce.
A ten-year-old pizza-making robot company, Picnic, went bankrupt despite raising over $20 million. In the same stretch, an assessment of Tesla's humanoid found zero units doing useful work, and Korean unions blocked a 25,000-robot deployment until a labor agreement is reached. Capability is rarely the thing that kills these projects — unit economics and worker resistance are, which is where due diligence should focus.
An independent German lab published the first real safety test of a popular humanoid — and it failed. The Unitree G1 hit with twice the force that causes a person pain and shipped with a hackable Bluetooth flaw. Vendor safety claims can no longer be taken at face value; insist on third-party testing before anything operates near people.
The data problem, not the hardware, is now the bottleneck. A survey of 700-plus robotics teams found only about a third have reached production, with poor-quality data the number-one cause of failure. Budget for data cleanup and integration, not just the robots themselves.

Coming Up

New EU and UK machinery safety rules covering AI-based and self-learning systems land in 2027. Compliance is becoming a selling point for equipment makers. If you buy or build automated equipment, start asking suppliers now how they will meet the cybersecurity and AI-safety requirements.
China and South Korea are building national robotics infrastructure that the West has no answer to. China has registered 28,000 humanoids on a state ID system; South Korea is funding 500 AI factories. Expect cheaper, faster-improving Asian robots — factor a widening cost gap into any multi-year automation plan.
Humanoid prices are projected to fall from roughly $115,000 today toward $37,000 by 2030. The math that fails today will start to work within this planning horizon. Identify your two or three most repetitive, hard-to-staff manual tasks now so you are ready to pilot when the price lands.

What's Hard About This

Narrow tasks pay; broad ambition fails. Tightly-scoped automation succeeds roughly half the time; general AI projects fail about 70 percent of the time, and 77 percent of automotive vision pilots never reach full production. The discipline to keep the scope small is the single biggest predictor of return.
The savings are real but rarely show up automatically. UK firm case studies found promised productivity gains often did not materialize, and most companies that deployed AI maintenance saw no change in downtime. The robot is the easy part; redesigning the work around it is where the value is won or lost.
Safety, security, and worker trust now gate deployment. Recalls, hackable consumer robots, and union resistance are slowing rollouts as much as any technical limit. Plan for independent safety testing and genuine workforce engagement, not just a purchase order.

Go deeper: the full Physical AI & Robotics briefing — the longer analytical write-up, plus every practice we track in this domain with its maturity rating, the tools to consider, and the evidence behind our assessment.