Open Models Almanac
Methodstraining objectivelanguage modelling objective

specific method · filed under training objective

Autoregressive discrete-action prediction

Uses autoregressive next-token prediction to generate discrete robot action or trajectory tokens, including FAST-tokenized action representations where specified.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See language modelling objective for the methods filed beside it.

Also called FAST encoded action tokens, autoregressive VLA.

sources
3
models
2
lab adopt it
1
strongest
core

Documented in

Physical Intelligence

Evidence

3 quotes from 3 sources, one card per model, each opening with its strongest.

How sources treat it
evaluated 1used 1core 1

π0.5

Physical Intelligenceadopted2 sources

the π₀-FAST model, an autoregressive VLA, based on the FAST action tokenizer

coremodel architecturein π₀-FASTPhysical Intelligence
1 more from this model

It is trained as a standard auto-regressive transformer, performing next-token prediction of text, object locations, and FAST encoded action tokens.

usedunclearin π0.5
↑