Polymorph/Physical AI, ExplainedGlossary

Physical AI glossary

What is Benchmark?

Benchmark: a fixed set of tasks and a scoring rule that every model is run on, so results can be compared. Useful for direction, easy to over-fit.

In the guide: 2.4 Imitation learning and VLAs went mainstream

The debate: The report argues that beyond roughly 7B parameters, curation beats scale: a well-curated 500-demonstration fine-tune of a 7B model outperforms a poorly curated fine-tune of a 70B model on most manipulation benchmarks. If it holds, competitive advantage lies in collecting and cleaning task data rather than in training ever larger models, though benchmark results do not always carry over to real deployments.

Read 2.4 in context →

In the guide: 4.10 Evaluation

Read 4.10 in context →

In the guide: 4.11 What the benchmarks show

Most robot results are measured on a team's own tasks, so two results are rarely comparable. A shared benchmark runs every model on the same tasks under the same scoring. One of the most complete in 2026 is RoboDojo: 42 manipulation tasks in simulation and 18 on real robots. Instead of one success rate, it scores five separate abilities:

Read 4.11 in context →

All 54 terms in the glossary · Read Physical AI, Explained