AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

M2Tok: Multi-head Multi-codebook Discrete Action Tokenization for Vision-Language-Action Models

Chunpu Xu, Zhixuan Liang, Yuhao Zhang, Chi-Min Chan, Jiashuo Wang, Yang Xiao, Mengkang Hu, Xiaokang Yang, Yao Mu

ECCV 2026 · Vision-language and multimodal › Robot action and navigation · approach: Vision-language-action foundation models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add multi-head decomposition (vs single head) (ablation)M2Tok · RoboTwinaverage success rate: 0.22 -> 0.51Improved
add multiple codebooks (vs single codebook) (ablation)M2Tok · RoboTwinaverage success rate: 0.34 -> 0.51Improved
add conversion module (vs direct embedding layer) (ablation)M2Tok · RoboTwinaverage success rate: 0.42 -> 0.51Improved
add multi-head to Move Pillbottle Pad (ablation)M2Tok · RoboTwin Move Pillbottle Padsuccess rate: 0.00 -> 0.33Improved
increase number of heads h up to 8 (ablation)M2Tok · RoboTwinsuccess rateImproved
increase number of heads h beyond 8 (ablation)M2Tok · RoboTwinsuccess rateNo clear effect
use M2Tok instead of VQ-VLA tokenizerVLA (Qwen2.5-0.5B) · RoboTwinaverage success rate: 0.45 -> 0.51Improved
use M2Tok instead of VQ-VLA tokenizerVLA (Qwen2.5-0.5B) · Simpler-Envaverage success rate: 0.21 -> 0.28Improved
use M2Tok instead of VQ-VLA tokenizerVLA (Qwen2.5-0.5B) · Real-world zero-shotaverage success rate: 0.28 -> 0.33Improved
use M2Tok instead of VQ-VLA on Stack Green on YellowVLA (Qwen2.5-0.5B) · Simpler-Envsuccess rate: 0.21 -> 0.08Hurt
use M2Tok instead of Binning tokenizer for inference speedVLA (Qwen2.5-0.5B) · RoboTwin latencyspeed (Hz): 2.0 -> 8.2Improved
use vLLM inference engineM2Tok VLA · RoboTwin latencyspeed (Hz): 8.2 -> 56.2Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy