ActPrune: Action-Grounded Visual Token Pruning for Efficient Vision-Language-Action Models
ActPrune: Action-Grounded Visual Token Pruning for Efficient Vision-Language-Action Models
This research presents ActPrune, a training-free method that grounds visual token pruning in geometric priors derived from robot actions to reduce inference costs in vision-language-action (VLA) models. By leveraging action chunks predicted at the previous control step, the method retains relevant patches in fixed third-person views while pruning unchanged regions in wrist views after compensating for camera motion. Experiments on simulation environments (LIBERO and VLABench) and four real-robot tasks demonstrate up to 1.61x speedup and 59.5% FLOPs reduction with negligible degradation in success rates.
The paper introduces a training-free token pruning method grounded in geometric priors from robot actions to reduce inference costs in vision-language-action models, achieving up to 1.61x speedup and 59.5% FLOPs reduction with negligible success rate degradation across simulation and real-robot tasks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
speedup
≤1.61
View original evidence
trate that \ours achieves up to 1.61 $\times$ speedup and 59.5\% FLOPs reduction with negligible degradation in success rates.
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
trate that \ours achieves up to 1.61 $\times$ speedup and 59.5\% FLOPs reduction with negligible degradation in success rates.
Open source S5
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 demons
Open source S4
re 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
Open source S3
arXiv:2611.08827v1 Announce Type: new Abstract: 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 enti
Open source S2
Source：arXiv Robotics — research abstracts · arxiv.org