AI Paper Atlas

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

MergeVLA: Cross-Skill Model Merging Toward a Generalist Vision-Language-Action Agent

Yuxia Fu, Zhizhen Zhang, Yuqi Zhang, Zijian Wang, Zi Huang, Yadan Luo

CVPR 2026 · Vision-language and multimodal › Robot action and navigation · approach: Vision-language-action foundation models · 15 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 task-specific LoRA mask to TA merge (ablation)OpenVLA TA merge (all) · LIBERO avgsuccess rate: 0.0 -> 62.4Improved
add task mask to VLA-Adapter merge (ablation)VLA-Adapter TA · LIBERO avgsuccess rate: 0.0 -> 0.0No clear effect
leave final action expert block unmerged (expert head) (ablation)VLA-Adapter TA + mask · LIBERO avgsuccess rate: 0.0 -> 23.1Improved
remove self-attention and use sigmoid gate in action expert (MergeVLA architecture) (ablation)VLA-Adapter TA + mask except head · LIBERO avgsuccess rate: 23.1 -> 89.7Improved
MergeVLA architecture single-task finetuning vs VLA-AdapterMergeVLA · LIBERO avgsuccess rate: 98.5 -> 96.7Hurt
MergeVLA architecture robustness vs VLA-Adapter (single-task)MergeVLA · LIBERO-Plus avgsuccess rate: 59.0 -> 72.4Improved
route with value subspace only instead of key (ablation)test-time task router · LIBERO avgsuccess rate: 53.6 -> 89.7Improved
route with value only instead of K&V (ablation)test-time task router · LIBERO avgsuccess rate: 65.1 -> 89.7Improved
use TIES instead of TA for mergingMergeVLA · LIBERO avgsuccess rate: 89.7 -> 90.2Improved
use TIES instead of TA for mergingMergeVLA head L-1..L · RoboTwin Setting Aavg success: 78.7 -> 88.7Improved
extend expert head from last 2 to last 3 blocks with TIES (ablation)MergeVLA · RoboTwin Setting Bavg success: 59.3 -> 70.7Improved
extend expert head from last 2 to last 3 blocks with TA (ablation)MergeVLA · RoboTwin Setting Bavg success: 48.7 -> 49.3No 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