mlsanityData to 7 Oct 2026

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

arXiv 2505.21200 ↗First posted 27 May 2025cs.CVcited by 30

On the map: Vision-language-action models → Efficient VLA inference and adaptation

Abstract (arXiv)

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.

Key facts. Cited by 30 papers in the map (data through 7 Oct 2026). Placed in Efficient VLA inference and adaptation, one of 13 directions in Vision-language-action models. Builds on 29 of its 48 references that are in the map. First posted on arXiv 27 May 2025.

Builds on

29 of 48 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025PD-VLA: Accelerating Vision-Language-Action Model Integrated with Action Chunking via Parallel Decoding4 Mar
  2. 2025VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching4 Feb
  3. 2025FAST: Efficient Action Tokenization for Vision-Language-Action Models16 Jan
  4. 2024Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Control2 Dec
  5. 2024Diffusion Transformer Policy21 Oct

Other references · 24

  1. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  2. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  3. 2022Do As I Can, Not As I Say: Grounding Language in Robotic Affordancescited by 2,302
  4. 2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyondcited by 2,179
  5. 2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,007
  6. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 844
  7. 2024DROID: A Large-Scale In-The-Wild Robot Manipulation Datasetcited by 801
  8. 2022R3M: A Universal Visual Representation for Robot Manipulationcited by 678
  9. 2024RDT-1B: a Diffusion Foundation Model for Bimanual Manipulationcited by 670
  10. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  11. 2024CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulationcited by 411
  12. 2024SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inferencecited by 378
12 more
  1. 2023Vision-Language Foundation Models as Effective Robot Imitatorscited by 344
  2. 2024VisionZip: Longer is Better but Not Necessary in Vision Language Modelscited by 298
  3. 2024GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulationcited by 290
  4. 2023RH20T: A Comprehensive Robotic Dataset for Learning Diverse Skills in One-Shotcited by 173
  5. 2025LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Tokencited by 164
  6. 2024RoboMamba: Efficient Vision-Language-Action Model for Robotic Reasoning and Manipulationcited by 102
  7. 2024Manipulate-Anything: Automating Real-World Robots using Vision-Language Modelscited by 82
  8. 2024Multi-Stage Vision Token Dropping: Towards Efficient Multimodal Large Language Modelcited by 67
  9. 2024Diffusion-VLA: Generalizable and Interpretable Robot Foundation Model via Self-Generated Reasoningcited by 57
  10. 2024Bidirectional Decoding: Improving Action Chunking via Guided Test-Time Samplingcited by 46
  11. 2024Inference Optimal VLMs Need Fewer Visual Tokens and More Parameterscited by 25
  12. 2024DNAct: Diffusion Guided Multi-Task 3D Policy Learningcited by 22

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 8; 2026: 22.

30 papers.

Sort:

By area:

  1. 2025Large VLM-based Vision-Language-Action Models for Robotic Manipulation: A Surveycited by 66
  2. 2025The Better You Learn, The Smarter You Prune: Towards Efficient Vision-language-action Models via Differentiable Token Pruningcited by 34
  3. 2025A Survey on Efficient Vision-Language-Action Modelscited by 34
  4. 2025Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulationcited by 32
  5. 2025Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Surveycited by 25
  6. 2025RLRC: Reinforcement Learning-based Recovery for Compressed Vision-Language-Action Modelscited by 20
  7. 2026GeoSem-WAM: Geometry- and Semantic-Aware World Action Modelscited by 8
  8. 2026AC^2-VLA: Action-Context-Aware Adaptive Computation in Vision-Language-Action Models for Efficient Robotic Manipulationcited by 6
  9. 2026TIDAL: Temporally Interleaved Diffusion and Action Loop for High-Frequency VLA Controlcited by 5
  10. 2026Efficient Long-Horizon Vision-Language-Action Models via Static-Dynamic Disentanglementcited by 5
  11. 2025ActDistill: General Action-Guided Self-Derived Distillation for Efficient Vision-Language-Action Modelscited by 4
  12. 2026OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelismcited by 4
18 more
  1. 2025Revisiting Multimodal KV Cache Compression: A Frequency-Domain-Guided Outlier-KV-Aware Approachcited by 3
  2. 2026AttenA+: Rectifying Action Inequality in Robotic Foundation Modelscited by 3
  3. 2026ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Modelscited by 2
  4. 2026Revisiting Parameter Redundancy in Vision-Language-Action Models: Insights from VLM-to-VLA Adaptationcited by 2
  5. 2026DepthCache: Depth-Guided Training-Free Visual Token Merging for Vision-Language-Action Model Inferencecited by 1
  6. 2026Beyond Attention Magnitude: Leveraging Inter-layer Rank Consistency for Efficient Vision-Language-Action Modelscited by 1
  7. 2026Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Modelscited by 1
  8. 2026UniFS: Unified Fast-to-Slow Hierarchical Architecture for Vision-Language-Action Modelscited by 1
  9. 2026Reducing Temporal Redundancy for Efficient Vision-Language-Action Inferencecited by 1
  10. 2026VLA-InfoEntropy: A Training-Free Vision-Attention Information Entropy Approach for Vision-Language-Action Models Inference Acceleration and Successcited by 0
  11. 20262D or 3D: Who Governs Salience in VLA Models? -- Tri-Stage Token Pruning Framework with Modality Salience Awarenesscited by 0
  12. 2026Improving Robotic Imitation Learning via Trajectory Standardizationcited by 0
  13. 2026FlashDrive: Flash Vision-Language-Action Inference for Autonomous Drivingcited by 0
  14. 2026UniMPA: A Unified Memory-Prediction-Action Model via Action-Grounded Transition Modelingcited by 0
  15. 2026VLA-ULAP: Interleaving Cloud VLA Calls with Ultra-Lightweight Local Action Prediction at the Edgecited by 0
  16. 2026Catch Me If You Can: Real-Time Feedback Denoising for Responsive VLAscited by 0
  17. 2026LLMs are General Asynchronous Agentscited by 0
  18. 2026VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Token Expand-Merge: Training-Free Token Compression for Vision-Language-Action Models
  2. 2026FlashVLA: Streaming Action Decoding for Fast and Asynchronous VLA Inference
  3. 2026Finetuning Vision-Language-Action Models Requires Fewer Layers Than You Think
  4. 2025FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via Neural Action Tokenization
  5. 2026A1: A Fully Transparent Open-Source, Adaptive and Efficient Truncated Vision-Language-Action Model
  6. 2026Learning to Accelerate Vision-Language-Action Models through Adaptive Visual Token Caching
  7. 2025Bridging the Semantic-Action Gap in Visual Token Pruning for Efficient VLA Inference
  8. 2025EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models