Open Models Almanac
Methodsdata curationtokenization

single source · data curation

FAST+

FAST+ is a universal robot action tokenizer trained on diverse real robot action trajectories.1 It is based on FAST, which uses the discrete cosine transform and byte-pair encoding to compress continuous robotic action signals.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. Based on FAST, we develop FAST+, a universal robot action tokenizer, trained on 1M real robot action trajectories that cover a large diversity of robot embodiments, action spaces and control frequencies — FAST: Efficient Action Tokenization for Vision-Language-Action Models Thanks: ∗: Core contributors Correspondence to:…
  2. FAST uses the discrete cosine transform (DCT) followed by byte-pair encoding (BPE) to compress action chunks, leading to significantly better compression than existing action tokenizers across a range of robotics domains — FAST: Efficient Action Tokenization for Vision-Language-Action Models Thanks: ∗: Core contributors Correspondence to:…

Also called FAST+ universal action tokenizer, universal action tokenizer, FAST+.

Filed under data curation › tokenization