Home › Hard problems › Physical AI, robots and autonomy

Physical AI, robots and autonomy

Too little robot training data

Solo: strongOne person can run a real test of this in about 8 weeks.

the web: trillions of wordsrecorded robot actions: very fewlab Alab Blab Cand every lab uses its own formatvs
Language models learned from trillions of words, but robots have only a few recorded actions to learn from, in formats that do not match.

Why it is still open

The web has trillions of words but very few recorded robot actions, and every lab uses its own format.

An experiment one person can run in 8 weeks

Build a checker that scores public robot datasets on quality (missing frames, mislabelled actions, sensor drift) and publish the results.

Verdict: Solo: strong

A data quality problem, which one careful person can measure.

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.

PreviousRobot hands that cope with new objectsNextThe gap between simulation and reality

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