FAST
Summary
FAST is a frequency-space action tokenizer for compressing robot action chunks into tokens usable by autoregressive VLAs.
Role In The Wiki
FAST is the main local caveat to a simple diffusion/flow narrative: better action tokenization keeps classical next-token VLA training competitive for some control-input trajectories.
Evidence
Official Artifacts
Relation To Foundation TSFM Agenda
Use the source-level agenda mapping in fast-2025 rather than duplicating verdict rows here.
At the entity level, FAST is the main local caveat to a simple diffusion/flow narrative: better action tokenization keeps classical next-token VLA training competitive for some control-input trajectories. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.