RESEARCH PAPER

FAST: Efficient Action Tokenization for Vision-Language-Action Models

Karl Pertsch; Kyle Stachowicz; Brian Ichter; Danny Driess; Suraj Nair; Quan Vuong; Oier Mees; Chelsea Finn; Sergey Levine

Classification

View four quadrants
Major category
Components of WAMs
Architecture
Not applicable
Prediction paradigm
Not applicable
Source review status
Verified from primary sources

Category review. FAST explicitly defines reusable action tokenization: normalize continuous chunks, apply temporal DCT, quantize coefficients and compress with BPE, then invert the representation for execution. FAST+ shares the tokenizer across embodiments. This is an action representation component. Reading evidence

AT A GLANCE

Contribution

A contribution summary has not been added yet.

Abstract

An abstract has not been added yet.

Affiliations

Not listed in the collection.

BibTeX