Skip to content
RobothecaBeta
Log in

X-Tokenizer

  • AI model
  • Humanoids and robot learning

A residual vector-quantization tokenizer that turns robot manipulation action sequences into short discrete token sequences and decodes them back, trained across 18 robot embodiments.

Made by X Square Robot, China (checked on github.com/X-Square-Robot (opens in a new tab)).

Source on GitHub: github.com/X-Square-Robot/X-Tokenizer (opens in a new tab). Facts read at commit 9733990 (opens in a new tab).

Licenses

Code
Apache-2.0 (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit 9733990 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Compresses action sequences by a factor of 4, with a codebook of 2048 codes across 4 residual quantizers per latent step.
  • Uses a 26-dimensional flat action layout covering bimanual end-effectors, chassis, lift and head.
  • Accepts any chunk length from 8 to 64 at inference without reloading the checkpoint.
  • Offers two token layouts, time-major and quantizer-major, for step-by-step or depth-first autoregressive consumers.
  • Requires per-robot statistics computed from your own data, or a provided statistics file, for normalization.
  • Installs with pip from the repository; the released checkpoint is downloaded from Hugging Face.

Written by Robotheca from the project's README (opens in a new tab) at commit 9733990.

Getting started

Difficulty
Advanced

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
18 Jun 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots that can run it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.