Pith. sign in

REVIEW 17 cited by

Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2505.21200 v1 pith:6AF7F2KU submitted 2025-05-27 cs.CV

Xudong Tan , Yaoxin Yang , Peng Ye , Jialin Zheng , Bizhe Bai , Xinyi Wang , Jia Hao , Tao Chen This is my paper
classification cs.CV
keywords actionflashvlainferencemodelsreuseacrosshighredundancy
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Vision-Language-Action (VLA) models have emerged as a powerful paradigm for general-purpose robot control through natural language instructions. However, their high inference cost-stemming from large-scale token computation and autoregressive decoding-poses significant challenges for real-time deployment and edge applications. While prior work has primarily focused on architectural optimization, we take a different perspective by identifying a dual form of redundancy in VLA models: (i) high similarity across consecutive action steps, and (ii) substantial redundancy in visual tokens. Motivated by these observations, we propose FlashVLA, the first training-free and plug-and-play acceleration framework that enables action reuse in VLA models. FlashVLA improves inference efficiency through a token-aware action reuse mechanism that avoids redundant decoding across stable action steps, and an information-guided visual token selection strategy that prunes low-contribution tokens. Extensive experiments on the LIBERO benchmark show that FlashVLA reduces FLOPs by 55.7% and latency by 36.0%, with only a 0.7% drop in task success rate. These results demonstrate the effectiveness of FlashVLA in enabling lightweight, low-latency VLA inference without retraining.

Discussion (0). Sign in to comment.

Forward citations

Cited by 17 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Revisiting Parameter Redundancy in Vision-Language-Action Models: Insights from VLM-to-VLA Adaptation

    cs.RO 2026-06 unverdicted novelty 7.0 of 10

    VLA models from VLM adaptation can be pruned 12-30% via multi-module joint scheme based on divergence signals while keeping ~90% performance on LIBERO without post-pruning recovery, unlike standard criteria that collapse.

  2. Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?

    cs.RO 2026-06 accept novelty 7.0 of 10

    VLA language backbones show high redundancy on manipulation benchmarks, with half the LLM blocks removable and even two blocks sufficient to recover baseline performance after fine-tuning, unlike vision and action pathways.

  3. Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models

    cs.RO 2026-05 unverdicted novelty 7.0 of 10

    Pace-and-Path Correction decomposes a quadratic cost minimization into orthogonal pace and path channels to correct chunked actions in VLA models, raising success rates by up to 28.8% in dynamic settings.

  4. KERV: Kinematic-Rectified Speculative Decoding for Embodied VLA Models

    cs.RO 2026-03 unverdicted novelty 7.0 of 10

    KERV integrates kinematic Kalman Filter predictions with speculative decoding in VLA models to achieve 27-37% faster inference while maintaining nearly the same task success rates.

  5. Improving Robotic Imitation Learning via Trajectory Standardization

    cs.RO 2026-06 unverdicted novelty 6.0 of 10

    ISR resamples trajectories to equal information spacing via velocity/acceleration norms on a Riemannian manifold, raising imitation learning success rates by ~25% over time-uniform downsampling on three real-world man...

  6. UniFS: Unified Fast-to-Slow Hierarchical Architecture for Vision-Language-Action Models

    cs.RO 2026-06 unverdicted novelty 6.0 of 10

    UniFS achieves 98.3% success on LIBERO with 2.1x lower latency than prior fast-slow VLA models by stratifying VLM layer update frequencies, inverting latent interactions, and applying multi-level supervision.

  7. Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Models

    cs.RO 2026-05 unverdicted novelty 6.0 of 10

    Pace-and-Path Correction is a closed-form inference-time operator that decomposes a quadratic cost minimization into orthogonal pace compression and path offset channels to correct dynamics-blindness in chunked-action...

  8. VLA-InfoEntropy: A Training-Free Vision-Attention Information Entropy Approach for Vision-Language-Action Models Inference Acceleration and Success

    cs.CV 2026-04 unverdicted novelty 6.0 of 10

    VLA-InfoEntropy accelerates Vision-Language-Action model inference by using visual entropy, attention entropy, and timestep cues to prune redundant tokens while preserving task-critical content.

  9. OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelism

    cs.RO 2026-03 unverdicted novelty 6.0 of 10

    OxyGen unifies KV cache management in MoT VLAs to enable cross-task KV sharing and cross-frame continuous batching, delivering up to 3.7x speedup with 200+ tokens/s language and 70 Hz action on on-device platforms.

  10. TIDAL: Temporally Interleaved Diffusion and Action Loop for High-Frequency VLA Control

    cs.RO 2026-01 conditional novelty 6.0 of 10

    TIDAL raises VLA control feedback from ~2.4 Hz to ~9 Hz by caching semantic intent and interleaving one-step flow generation with execution, doubling dynamic interception success in simulation.

  11. ActDistill: General Action-Guided Self-Derived Distillation for Efficient Vision-Language-Action Models

    cs.CV 2025-11 conditional novelty 6.0 of 10

    ActDistill transfers action knowledge from heavy VLA teacher models to lightweight students via graph-encapsulated hierarchies and action-guided dynamic routing, delivering over 50% computation reduction and 1.67x spe...

  12. GeoSem-WAM: Geometry- and Semantic-Aware World Action Models

    cs.RO 2026-06 unverdicted novelty 5.0 of 10

    GeoSem-WAM adds geometric and semantic auxiliary prediction tasks to World Action Models during training to improve latent representations and action prediction accuracy while keeping inference efficient by avoiding e...

  13. ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models

    cs.RO 2026-05 unverdicted novelty 5.0 of 10

    ElegantVLA accelerates VLA models up to 3.77x by dynamically scheduling compute across vision, language, and action components without retraining the base model.

  14. AttenA+: Rectifying Action Inequality in Robotic Foundation Models

    cs.RO 2026-05 unverdicted novelty 5.0 of 10

    AttenA+ applies velocity-driven action attention to reweight training objectives toward kinematically critical low-velocity segments, yielding small benchmark gains on Libero and RoboTwin without added parameters.

  15. Large VLM-based Vision-Language-Action Models for Robotic Manipulation: A Survey

    cs.RO 2025-08 unverdicted novelty 5.0 of 10

    This survey organizes large VLM-based VLA models for robotic manipulation into monolithic and hierarchical paradigms, reviews their integrations and datasets, and outlines future directions.

  16. AttenA+: Rectifying Action Inequality in Robotic Foundation Models

    cs.RO 2026-05 unverdicted novelty 4.0 of 10

    AttenA+ reweights action training objectives in VLA and WAM models via inverse velocity attention to prioritize kinematically critical segments, yielding small benchmark gains.

  17. Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey

    cs.RO 2025-10 conditional novelty 4.0 of 10

    A survey that groups VLA efficiency techniques into four categories: model architecture, perception features, action generation, and training/inference strategies.

Pith tools