Home › Hard problems › Physical AI, robots and autonomy

Physical AI, robots and autonomy

The gap between simulation and reality

Solo: possibleOne person can test a narrow slice of it.

In simulationeven light,exact frictionthe same policyIn the real worldslippery surface, glare, worn fingerstrained here
A robot that learned in a perfect simulator meets real friction, light and wear, and fails.

Why it is still open

Robots trained in simulation fail on real friction, light and wear.

An experiment one person can run in 8 weeks

In a free simulator (MuJoCo, Isaac Sim), measure how a trained policy degrades as you randomise physics.

Verdict: Solo: possible

Simulation-only results are a partial answer.

The solo test behind the verdict

  1. Can it be measured? There is a number that shows you did better, such as an error rate, a catch rate or a cost per answer.
  2. Is it cheap to try? It runs on a laptop, free open models and open data, or under about ₹50,000 of rented GPU time.
  3. Is there a public yardstick? A benchmark or dataset others use, so your result can be compared and believed.
  4. Would a small win get noticed? A lab, a funder or a customer would care about a 10% improvement, not only a full solution.

Four yeses means Solo: strong. Two or three means Solo: possible. Fewer means Follow, don't build.

PreviousToo little robot training dataNextProving an autonomous system is safe

All 79 hard problems · Working on this, or know of a result we missed? basant@cleandataengineering.ai