AI Paper Atlas

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

See What Matters: Differentiable Grid Sample Pruning for Generalizable Vision-Language-Action Model

Yixu Feng, Zinan Zhao, Yanxiang Ma, Chenghao Xia, Chengbin Du, Yunke Wang, Chang Xu

ICML 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
replace post-training discrete pruning (FastV, SparseVLM) with jointly trained GridS16π0 · LIBERO averagesuccess rate (%): 94.4 -> 96.0Improved
reduce visual tokens from 256 to 4 with GridS (ablation)π0.5 · LIBERO averagesuccess rate (%): 96.7 -> 96.7No clear effect
reduce visual tokens to 16 with GridSπ0 · LIBERO FLOPsFLOPs (G): 216.01 -> 51.65Improved
reduce GridS token budget to K=4 (extreme sparsity) (ablation)π0.5 · LIBERO-Spatial and Longsuccess rate (%): Long 95.2 -> 92.8Hurt
reduce GridS token budget to K=4 on Object suite (ablation)π0 · LIBERO-Objectsuccess rate (%): 99.2 -> 99.4Improved
use GridS with 16 tokens in place of 256 on ALOHAπ0 · ALOHA Insertion (Human)accuracy (%): 56.7 -> 64.2Improved
use GridS with 16 tokens on ALOHA Insertion (Scripted)π0 · ALOHA Insertion (Scripted)accuracy (%): 91.4 -> 86.9Hurt
add GridS (16 tokens) to SmolVLA on real robotSmolVLA · Real-world Stack Cubessuccess rate (%): 9.5 -> 61.9Improved
vary sampled token count K from 4 to 8 to 16 to 32 (ablation)GridS on SmolVLA · Real-world Stack Cubessuccess rate (%): 0.0 / 28.5 / 61.9 / 19.0Mixed
add GridS (K=16) for OOD evaluationSmolVLA · Real-world Stack Cubes OOD (21 scenarios)OOD success rate (%): 0.0 -> 38.1Improved
add GridS to trainingπ0 · training time per step (batch 128)time (s/step): 14.32 -> 4.21Improved
add GridS (K=16) under wrist-only inference with token maskingπ0 on LIBERO · LIBERO wrist-only (Long)accuracy: 0.30 -> 0.70Improved

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