AI Paper Atlas

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

MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation

Rongyu Zhang, Menghang Dong, Yuan Zhang, Liang Heng, Xiaowei Chi, Gaole Dai, Li Du, Dan Wang, Yuan Du, Shanghang Zhang

AAAI 2026 · Vision-language and multimodal › Robot action and navigation · approach: Vision-language-action foundation models · 78 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
use STAR router instead of a linear router (ablation)MoLe-OpenVLA (50% layer skip) · RLBench (OpenVLA)mean success rate (%): 42.6 -> 45.6Improved
skip 50% of layers with a linear router (no distillation) (ablation)OpenVLA · RLBenchmean success rate (%): 46.7 -> 42.6Hurt
add CogKD with MSE loss (ablation)MoLe-OpenVLA with STAR · RLBenchmean success rate (%): 45.6 -> 51.4Improved
add CogKD with forward KL loss (ablation)MoLe-OpenVLA with STAR · RLBenchmean success rate (%): 45.6 -> 49.6Improved
add CogKD with reverse KL loss (ablation)MoLe-OpenVLA with STAR · RLBenchmean success rate (%): 45.6 -> 53.8Improved
combine MSE and reverse KL in CogKD (ablation)MoLe-OpenVLA with STAR · RLBenchmean success rate (%): 53.8 -> 56.4Improved
apply MoLe layer-skipping to OpenVLAOpenVLA · RLBenchmean success rate (%): 46.7 -> 56.4Improved
apply MoLe layer-skipping to pi0pi0 · RLBenchmean success rate (%): 50.0 -> 55.4Improved
apply MoLe layer-skipping to CogActCogAct · RLBenchmean success rate (%): 57.1 -> 60.2Improved
random layer skipping instead of learned routingCogAct · RLBenchmean success rate (%): 57.1 -> 51.2Hurt
token-wise MoD routing instead of MoLe layer routingCogAct · RLBenchmean success rate (%): 57.1 -> 56.5No clear effect
early-exit DeeR instead of MoLeCogAct · RLBenchmean success rate (%): 58.8 -> 60.2No clear effect

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