OpenVLA / OpenVLA-OFT
OpenVLA project (Stanford et al.)we discretize each dimension of the robot actions separately into one of 256 bins
implementation detail · filed under data curation
Robot actions are tokenized by discretizing each dimension separately into bins; the cited implementation uses 256 bins.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See tokenization for the methods filed beside it.
Also called Per-dimension, per-timestep action binning, per-dimension, per-timestep binning scheme.
2 quotes from 2 sources, one card per model, each opening with its strongest.
we discretize each dimension of the robot actions separately into one of 256 bins
Prior robotic policies of this sort typically use naïve tokenization strategies based on a per-dimension, per-timestep binning scheme