[ Evidence and evaluation · no robots operated here ]

Physical AI

Learned models are being asked to act in the physical world, where the system’s own actions decide what it sees next and a mistake has consequences. These pages explain the methods — world models, vision-language-action policies, learning from demonstration, domain randomization, uncertainty and monitoring — and say plainly what each published result establishes.

This section is the learning and modelling companion to robotics evidence and evaluation, which owns the hardware, the evidence records and the safety boundaries. Where a subject belongs to both, robotics keeps the evidence record and these pages link to it rather than publishing a second version.

Start here

All 16 explanations

Run the fixture yourself

A deterministic one-dimensional perception–action loop with observation noise, actuation delay, a monitor and an intervention path. It reports autonomous success, assisted success, failure or abort — and splits monitor firings into those an injected disturbance explains and those it does not.

node --experimental-strip-types scripts/physical-ai-loop.ts run --actuationDelaySteps 3
node --experimental-strip-types scripts/physical-ai-loop.ts counterexamples

It is a simulation with no physics, contact, hardware or people. See what the fixture is not.

Related sections

Sources read for this section

Every performance figure on these pages is the original authors’ own reported result on their own evaluation, attributed as such. Maha has replicated none of them, operates no robot, and endorses no system or vendor.