← 研踪 PaperTrailarXiv:2611.08827 · v1
ActPrune: Action-Grounded Visual Token Pruning for Efficient Vision-Language-Action Models
首次提交 2026-11-06 08:32
版本更新 2026-11-06 08:32
雷达观测 2026-11-09 10:16
摘要
Vision-language-action (VLA) models perform versatile manipulation tasks from language instructions, but their inference cost limits how fast a robot can respond in closed-loop control. Visual token pruning reduces this cost; however, prior methods estimate token redundancy from encoded visual embeddings or attention, only after the visual encoder has processed the entire observation. We propose \ours, a training-free token pruning method that moves the pruning decision ahead of the vision encoder by grounding it in a geometric prior derived from the robot's own action. The action chunk predicted at the previous control step is already available before the current observation is encoded, and it specifies the end-effector and camera motion between consecutive observations. The same action affects the two standard camera views asymmetrically: in the fixed third-person view, \ours retains patches along the projected end-effector trajectory together with semantic anchors, while in the wrist view, it prunes patches that remain unchanged after compensating for the action-induced camera motion. Extensive experiments on two simulation environments (LIBERO and VLABench) and four real-robot tasks demonstrate that \ours achieves up to 1.61 $\times$ speedup and 59.5\% FLOPs reduction with negligible degradation in success rates.
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-11 16:35 · 最新观测版本
ActPrune: Action-Grounded Visual Token Pruning for Efficient Vision-Language-Action Models
Vision-language-action (VLA) models perform versatile manipulation tasks from language instructions, but their inference cost limits how fast a robot can respond in closed-loop control. Visual token pruning reduces this cost; however, prior methods estimate token redundancy from encoded visual embeddings or attention, only after the visual encoder has processed the entire observation. We propose ActPrune, a training-free token pruning method that moves the pruning decision ahead of the vision encoder by grounding it in a geometric prior derived from the robot's own action. The action chunk predicted at the previous control step is already available before the current observation is encoded, and it specifies the end-effector and camera motion between consecutive observations. The same action affects the two standard camera views asymmetrically: in the fixed third-person view, ActPrune retains patches along the projected end-effector trajectory together with semantic anchors, while in the wrist view, it prunes patches that remain unchanged after compensating for the action-induced camera motion. Extensive experiments on two simulation environments (LIBERO and VLABench) and four real-robot tasks demonstrate that ActPrune achieves up to 1.61$\times$ speedup and 59.5% FLOPs reduction with negligible degradation in success rates. The project site is http://mmai.ewha.ac.kr/actprune/.
查看此版本解读 →查看此版本原文 ↗v1 · 2026-11-06 08:32
ActPrune: Action-Grounded Visual Token Pruning for Efficient Vision-Language-Action Models
Vision-language-action (VLA) models perform versatile manipulation tasks from language instructions, but their inference cost limits how fast a robot can respond in closed-loop control. Visual token pruning reduces this cost; however, prior methods estimate token redundancy from encoded visual embeddings or attention, only after the visual encoder has processed the entire observation. We propose \ours, a training-free token pruning method that moves the pruning decision ahead of the vision encoder by grounding it in a geometric prior derived from the robot's own action. The action chunk predicted at the previous control step is already available before the current observation is encoded, and it specifies the end-effector and camera motion between consecutive observations. The same action affects the two standard camera views asymmetrically: in the fixed third-person view, \ours retains patches along the projected end-effector trajectory together with semantic anchors, while in the wrist view, it prunes patches that remain unchanged after compensating for the action-induced camera motion. Extensive experiments on two simulation environments (LIBERO and VLABench) and four real-robot tasks demonstrate that \ours achieves up to 1.61 $\times$ speedup and 59.5\% FLOPs reduction with negligible degradation in success rates.
查看此版本解读 →查看此版本原文 ↗