Research map / Vision-language and multimodal / Robot action and navigation
MoEActok: A MoE-based Action Tokenizer for Vision-Language-Action Models
Chunpu Xu, Zhixuan Liang, Tianshuo Yang, Chi-Min Chan, Yang Xiao, Jessie Wang, Xiaokang Yang, Yao Mu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| remove pre/post-quantization adapters (ablation) | MoEActok · RoboTwin (12 tasks) | average success rate: 0.45 -> 0.56 | Improved |
| add pre/post-quantization adapters (ablation) | MoEActok · Simpler-Env WidowX | average success rate: 0.17 -> 0.38 | Improved |
| add skill-aware VLA training with explicit skill prediction (ablation) | MoEActok · RoboTwin (12 tasks) | average success rate: 0.47 -> 0.56 | Improved |
| add skill-aware VLA training with explicit skill prediction (ablation) | MoEActok · Simpler-Env WidowX | average success rate: 0.30 -> 0.38 | Improved |
| increase number of expert quantizers from 1 to 4 (ablation) | MoEActok · RoboTwin (12 tasks) | average success rate: 0.50 -> 0.56 | Improved |
| increase number of expert quantizers from 1 to 4 (ablation) | MoEActok · Simpler-Env WidowX | average success rate: 0.26 -> 0.38 | Improved |
| skill-aware training on a single task (ablation) | MoEActok · RoboTwin Handover Mic | success rate: 0.99 -> 0.95 | No clear effect |
| use MoEActok instead of VQ-VLA tokenizer | VLA with Qwen2.5-0.5B · RoboTwin (12 tasks) | average success rate: 0.45 -> 0.56 | Improved |
| use MoEActok instead of VQ-VLA tokenizer | VLA with Qwen2.5-0.5B · Simpler-Env Stack Green on Yellow | success rate: 0.21 -> 0.13 | Hurt |
| use MoEActok instead of VQ-VLA tokenizer | VLA with Qwen2.5-0.5B · Real robot zero-shot (3 tasks) | average success rate: 0.28 -> 0.37 | Improved |
| use FAST tokenizer instead of binning | VLA with Qwen2.5-0.5B · RoboTwin (12 tasks) | average success rate: 0.24 -> 0.17 | Hurt |
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.
- VQ-VLA — VQ-VLA: Improving Vision-Language-Action Models via Scaling Vector-Quantized Action Tokenizers (ICCV 2025) · better on 5 of 7 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
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.91)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.90)
- VLA-OS: Structuring and Dissecting Planning Representations and Paradigms in Vision-Language-Action Models (NeurIPS 2025 · similarity 0.90)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
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.
