mlsanityData to 7 Oct 2026

VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching

arXiv 2502.02175 ↗First posted 4 Feb 2025cs.RO cs.CV cs.LGcited by 83

On the map: LLM reasoning → Spatial and embodied reasoning

also in 1 more
Abstract (arXiv)

Vision-Language-Action (VLA) models have demonstrated strong multi-modal reasoning capabilities, enabling direct action generation from visual perception and language instructions in an end-to-end manner. However, their substantial computational cost poses a challenge for real-time robotic control, where rapid decision-making is essential. This paper introduces VLA-Cache, a training-free inference acceleration method that reduces computational overhead by adaptively caching and reusing static visual tokens across frames. Exploiting the temporal continuity in robotic manipulation, VLA-Cache identifies minimally changed tokens between adjacent frames and reuses their cached key-value representations, thereby circumventing redundant computations. Additionally, to maintain action precision, VLA-Cache selectively re-computes task-relevant tokens that are environmentally sensitive, ensuring the fidelity of critical visual information. To further optimize efficiency, we introduce a layer adaptive token reusing strategy that dynamically adjusts the reuse ratio based on attention concentration across decoder layers, prioritizing critical tokens for recomputation. Extensive experiments on two simulation platforms (LIBERO and SIMPLER) and a real-world robotic system demonstrate that VLA-Cache achieves up to 1.7x speedup in CUDA latency and a 15% increase in control frequency, with negligible loss on task success rate. The code and videos can be found at our project page: https://vla-cache.github.io.

Key facts. Cited by 83 papers in the map (data through 7 Oct 2026). Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning. Builds on 35 of its 42 references that are in the map. First posted on arXiv 4 Feb 2025.

Builds on

35 of 42 references.

Closest ancestors · 2

Same direction, newest first

  1. 2024DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution4 Nov
  2. 2024ReKep: Spatio-Temporal Reasoning of Relational Keypoint Constraints for Robotic Manipulation3 Sep

Other references · 33

  1. 2017Attention Is All You Needcited by 59,932
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  3. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  4. 2023Visual Instruction Tuningcited by 7,792
  5. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  6. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  7. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 2,730
  8. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  9. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,928
  10. 2023Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardwarecited by 1,490
  11. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  12. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
21 more
  1. 2022Token Merging: Your ViT But Fastercited by 995
  2. 2023Open X-Embodiment: Robotic Learning Datasets and RT-X Modelscited by 849
  3. 2025FAST: Efficient Action Tokenization for Vision-Language-Action Modelscited by 643
  4. 2024An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Modelscited by 629
  5. 2023OpenFlamingo: An Open-Source Framework for Training Large Autoregressive Vision-Language Modelscited by 560
  6. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  7. 2024CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulationcited by 411
  8. 2024SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inferencecited by 378
  9. 2023Vision-Language Foundation Models as Effective Robot Imitatorscited by 344
  10. 2024TinyVLA: Towards Fast, Data-Efficient Vision-Language-Action Models for Robotic Manipulationcited by 331
  11. 2023RH20T: A Comprehensive Robotic Dataset for Learning Diverse Skills in One-Shotcited by 173
  12. 2024LLARVA: Vision-Action Instruction Tuning Enhances Robot Learningcited by 87
  13. 2024HiRT: Enhancing Robotic Control with Hierarchical Robot Transformerscited by 72
  14. 2023PuMer: Pruning and Merging Tokens for Efficient Vision Language Modelscited by 53
  15. 2024MADTP: Multimodal Alignment-Guided Dynamic Token Pruning for Accelerating Vision-Language Transformercited by 53
  16. 2024MoPE-CLIP: Structured Pruning for Efficient Vision-Language Models with Module-wise Pruning Error Metriccited by 30
  17. 2024Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Controlcited by 25
  18. 2024Efficient Track Anythingcited by 20
  19. 2023Invariant Causal Imitation Learning for Generalizable Policiescited by 15
  20. 2024Advancing Multimodal Large Language Models with Quantization-Aware Scale Learning for Efficient Adaptationcited by 15
  21. 2023Domain Adaptive Imitation Learning with Visual Observationcited by 11

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 16; 2026: 67.

83 papers.

Sort:

By area:

  1. 2025Vision-Language-Action Models for Robotics: A Review Towards Real-World Applicationscited by 82
  2. 2025SP-VLA: A Joint Model Scheduling and Token Pruning Approach for VLA Model Accelerationcited by 56
  3. 2025Pure Vision Language Action (VLA) Models: A Comprehensive Surveycited by 46
  4. 2025SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruningcited by 39
  5. 2025CEED-VLA: Consistency Vision-Language-Action Model with Early-Exit Decodingcited by 38
  6. 2025A Survey on Efficient Vision-Language-Action Modelscited by 34
  7. 2025Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulationcited by 32
  8. 2025Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Modelscited by 30
  9. 2025RLRC: Reinforcement Learning-based Recovery for Compressed Vision-Language-Action Modelscited by 20
  10. 2025Align-Then-stEer: Adapting the Vision-Language Action Models through Unified Latent Guidancecited by 18
  11. 2026Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoningcited by 17
  12. 2025SQAP-VLA: A Synergistic Quantization-Aware Pruning Framework for High-Performance Vision-Language-Action Modelscited by 15
38 more
  1. 2025Bridging the Semantic-Action Gap in Visual Token Pruning for Efficient VLA Inferencecited by 14
  2. 2026DyQ-VLA: Temporal-Dynamic-Aware Quantization for Embodied Vision-Language-Action Modelscited by 10
  3. 2026VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Modelcited by 9
  4. 2026TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAMcited by 9
  5. 2026A1: A Fully Transparent Open-Source, Adaptive and Efficient Truncated Vision-Language-Action Modelcited by 8
  6. 2026Shallow-π: Knowledge Distillation for Flow-based VLAscited by 7
  7. 2025Focusing on What Matters: Object-Agent-centric Tokenization for Vision Language Action modelscited by 6
  8. 2026AC^2-VLA: Action-Context-Aware Adaptive Computation in Vision-Language-Action Models for Efficient Robotic Manipulationcited by 6
  9. 2026StreamingVLA: Streaming Vision-Language-Action Model with Action Flow Matching and Adaptive Early Observationcited by 6
  10. 2026Characterizing Vision-Language-Action Models across XPUs: Constraints and Acceleration for On-Robot Deploymentcited by 6
  11. 2025Experiences from Benchmarking Vision-Language-Action Models for Robotic Manipulationcited by 5
  12. 2026TIDAL: Temporally Interleaved Diffusion and Action Loop for High-Frequency VLA Controlcited by 5
  13. 2026FocusVLA: Focused Visual Utilization for Vision-Language-Action Modelscited by 5
  14. 2026See What Matters: Differentiable Grid Sample Pruning for Generalizable Vision-Language-Action Modelcited by 5
  15. 2025EgoPrune: Efficient Token Pruning for Egomotion Video Reasoning in Embodied Agentcited by 4
  16. 2026PACT-WAM: Predicting Actions and Visual Foresight with Compact Temporal Encoding for Robot Manipulationcited by 4
  17. 2026OxyGen: Unified KV Cache Management for VLA Inference under Multi-Task Parallelismcited by 4
  18. 2026CF-VLA: Efficient Coarse-to-Fine Action Generation for Vision-Language-Action Policiescited by 4
  19. 2026Sentinel-VLA: A Metacognitive VLA Model with Active Status Monitoring for Dynamic Reasoning and Error Recoverycited by 4
  20. 2026Latent Bridge: Feature Delta Prediction for Efficient Dual-System Vision-Language-Action Model Inferencecited by 4
  21. 2026SAFE-Pruner: Semantic Attention-Guided Future-Aware Token Pruning for Efficient Vision-Language-Action Manipulationcited by 4
  22. 2025Robotic Manipulation via Imitation Learning: Taxonomy, Evolution, Benchmark, and Challengescited by 3
  23. 2026AttenA+: Rectifying Action Inequality in Robotic Foundation Modelscited by 3
  24. 2026Kamera: Unified Position-Invariant Multimodal KV Cache for Training-Free Reusecited by 3
  25. 2026Reflex: Real-Time VLA Control through Streaming Inferencecited by 3
  26. 2026FlashVLA: Streaming Action Decoding for Fast and Asynchronous VLA Inferencecited by 3
  27. 2026HiCI: Hierarchical Construction-Integration for Long-Context Attentioncited by 2
  28. 2026Seeing Realism from Simulation: Efficient Video Transfer for Vision-Language-Action Data Augmentationcited by 2
  29. 2026AURA: Action-Gated Memory for Robot Policies at Constant VRAMcited by 2
  30. 2026PolicyTrim: Boosting Intrinsic Policy Efficiency of Vision-Language-Action Modelscited by 2
  31. 2026Revisiting Parameter Redundancy in Vision-Language-Action Models: Insights from VLM-to-VLA Adaptationcited by 2
  32. 2026Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on Heterogeneous Robotscited by 2
  33. 2026ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinementcited by 2
  34. 2026The Gate, Not the Cache: Gate Provenance Bounds the Closed-Loop Reliability of Training-Free VLA Token Skippingcited by 2
  35. 2026Decoupling Vision, Language, and Action for Efficient Multi-Task Robot Policiescited by 2
  36. 2026DepthCache: Depth-Guided Training-Free Visual Token Merging for Vision-Language-Action Model Inferencecited by 1
  37. 2026Overcoming Dynamics-Blindness: Training-Free Pace-and-Path Correction for VLA Modelscited by 1
  38. 2026Learned Image Compression for Vision-Language-Action Modelscited by 1

50 most cited of 83 All 83 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Learning to Accelerate Vision-Language-Action Models through Adaptive Visual Token Caching
  2. 2026D2-VLA: Dual-Memory Dual-Frequency Vision-Language-Action Model For Long Dynamic Manipulation
  3. 2026Reducing Temporal Redundancy for Efficient Vision-Language-Action Inference
  4. 2026ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models
  5. 2026DySL-VLA: Efficient Vision-Language-Action Model Inference via Dynamic-Static Layer-Skipping for Robot Manipulation
  6. 2026Efficient Long-Horizon Vision-Language-Action Models via Static-Dynamic Disentanglement
  7. 2025AVA-VLA: Improving Vision-Language-Action models with Active Visual Attention
  8. 2026Restoring Linguistic Grounding in VLA Models via Train-Free Attention Recalibration