AI Paper Atlas

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

PARL-VLA: Pruning-Aware On-Policy Reinforcement Learning for Vision-Language-Action Model

ruiyan xu, Jiashu Lv, Sixu Lin, Ruixing Jin, Shuliang He, Guiliang Liu

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
train with pruning inside the RL loop (PARL-VLA High) rather than full-token RLinfOpenVLA-OFT + GRPO · LIBERO-Plus meansuccess rate (%): 70.2 -> 74.0Improved
train with pruning in the RL loop rather than pruning only at evaluation (FastV)OpenVLA-OFT + GRPO · LIBERO-Plus meansuccess rate (%): 67.0 -> 74.0Improved
prune only at evaluation (VLA-Pruner) on a full-token-trained policyRLinf · LIBERO-Plus meansuccess rate (%): 70.2 -> 66.5Hurt
add PARL-VLA pruning-aware trainingRLinf · LIBERO-Plus Lightsuccess rate (%): 62.7 -> 68.7Improved
add PARL-VLA pruning-aware trainingRLinf · LIBERO-Plus Goal suite meansuccess rate (%): 66.7 -> 73.5Improved
add PARL-VLA pruning-aware trainingRLinf · LIBERO in-distribution avgsuccess rate (%): 94.4 -> 94.9No clear effect
evaluation-only pruning (FastV) instead of full tokensRLinf · LIBERO in-distribution avgsuccess rate (%): 94.4 -> 92.4Hurt
replay the recorded rollout pruning configuration during actor updates (vs full tokens at update) (ablation)PARL-VLA · LIBERO training stabilityapprox. KL: KL near 0.01 vs above 1Improved
replay the rollout pruning configuration rather than using a different update budget (min(1,1.1k)) (ablation)PARL-VLA · LIBERO trainingsuccess rateImproved
add spectrum training over keep ratios vs fixed rho=0.40 (ablation)PARL-VLA · LIBERO-Plus avgsuccess rate (%): 71.1 -> 73.5Improved
add spectrum training vs fixed rho=0.40 (ablation)PARL-VLA · LIBERO-Plus worst-casesuccess rate (%): 69.2 -> 72.7Improved
add spectrum training vs fixed rho=0.30 at eval keep ratio 0.30 (ablation)PARL-VLA · LIBERO-Plus eval rho=0.30success rate (%): 63.9 -> 72.7Improved

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