mlsanityData to 7 Oct 2026

Efficient VLA inference and adaptation

237 papers4 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: action, vla, inference, vision-language-action, while

Key facts. Efficient VLA inference and adaptation: 237 papers on arXiv since 2024 (data through 7 Oct 2026). 79 of its 237 mapped papers (33%) cite “FAST: Efficient Action Tokenization for Vision-Language-Action Models” (2025, arXiv 2501.09747). 42 of its 237 mapped papers (18%) cite “EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models” (2025, arXiv 2506.10100). 32 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

32 new papers, 17 Sep – 7 Oct; 4 this week.

  1. This week · 4
  2. ActTune: Action-Aware Precision and GPU Operating-Point Adaptation for Energy-Efficient Vision-Language-Action Inference
  3. VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models also fits: Generalist VLA models and benchmarks
  4. StairVLA: Stage-Aware Hierarchical Action Generation for Vision-Language-Action Models
  5. VLA-ZO: Fast Zeroth-Order Adaptation for Vision-Language-Action Models
  6. Earlier · 28
  7. Beyond LLM Serving: Characterizing Vision-Language-Action Workloads for Embodied AI System Design
  8. FastOPD: On-Policy Distillation for Lightweight VLA Deployment also fits: Reinforcement learning and post-training
  9. CHASE-VLA: Post-Training Quantization Framework for Vision-Language-Action Models with Chunk-Aware Scale Estimation
  10. TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models also fits: Reinforcement learning and post-training
  11. Toward Real-Time VLAs: Stage-Aware Two-Step Flow Denoising and System-Level Evaluation
  12. Discrete Forcing: Infusing Discrete Guidance into Continuous Denoising for Few-Step Action Experts
All 32
  1. Spike-driven Vision-Language-Action Model also fits: Generalist VLA models and benchmarks
  2. DSDyn-VLA: A Dual-Stream Dynamic Manipulation Framework with Motion Perception, Future Awareness, and Realtime Correction also fits: World model-based VLA
  3. Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control
  4. Reactive Real-Time Flow Policies via Asynchronous Distribution Alignment also fits: Reinforcement learning and post-training
  5. Alignment-Guided Flow Transformer for Efficient Vision-Language-Action Policy Learning also fits: Generalist VLA models and benchmarks
  6. Rethinking Causal Action Tokenization with Conditional Annealing in Flow Matching
  7. ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning also fits: Generalist VLA models and benchmarks
  8. Adjoint Guidance Flow: Amortized Critic Guidance for VLA Policies also fits: Reinforcement learning and post-training
  9. Text-Vision Synergistic Token Caching: A Training-Free Framework for Efficient Vision-Language-Action Inference
  10. RAVEL: Asynchronous Rolling Inference for Flow-Based Vision-Language-Action Models
  11. Quantile Head for Vision-Language-Action Models also fits: Generalist VLA models and benchmarks
  12. DS-VLA: A Dendritic-inspired Vision-Language-Action Model for Robust Action Control
  13. Fast Plans, Faithful Actions: Closing the Planning-Execution Gap in Hierarchical Vision-Language-Action Models also fits: Reasoning and chain-of-thought in VLA
  14. Decoupled Early Exits for Task-Dependent Compute Allocation in Flow-Matching VLAs
  15. Beyond Reconstruction Error: Analytical and Data-Driven Action Tokenization for Autoregressive Vision-Language-Action Models also fits: VLA architectures and generalization
  16. FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding
  17. SmoLSTM: A Compact Vision-Language-Action Model with Recurrent Memory that Persists also fits: Reasoning and chain-of-thought in VLA
  18. StateMem: Single-State Residual Memory with Adaptive Inference for Vision-Language-Action Policies also fits: Reasoning and chain-of-thought in VLA
  19. Fewer Steps, Better Actions: Rethinking Flow-Matching Inference for VLA Policies also fits: VLA architectures and generalization
  20. Catch Me If You Can: Real-Time Feedback Denoising for Responsive VLAs
  21. GeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA Policies
  22. Recovering Aggressively Pruned Vision-Language-Action Models with Offline Hidden-State Distillation also fits: Reinforcement learning and post-training

Foundations

Most cited by papers here

  1. FAST: Efficient Action Tokenization for Vision-Language-Action Models2025cited by 79 here
  2. EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models2025cited by 42 here
  3. VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching2025cited by 38 here
  4. MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation2025cited by 31 here
  5. PD-VLA: Accelerating Vision-Language-Action Model Integrated with Action Chunking via Parallel Decoding2025cited by 28 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 168 here
  2. 2024Octo: An Open-Source Generalist Robot Policycited by 114 here
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 204 here
  4. 2024TinyVLA: Towards Fast, Data-Efficient Vision-Language-Action Models for Robotic Manipulationcited by 74 here
  5. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 177 here
  6. 2025FAST: Efficient Action Tokenization for Vision-Language-Action Modelscited by 79 here
  7. 2025SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Modelcited by 52 here
  8. 2025PD-VLA: Accelerating Vision-Language-Action Model Integrated with Action Chunking via Parallel Decodingcited by 28 here
  9. 2025EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Modelscited by 42 here
  10. 2025SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruningcited by 24 here
14 more it builds on, not in this chain
  1. 2024A Survey on Vision-Language-Action Models for Embodied AIcited by 21 here
  2. 2024CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulationcited by 63 here
  3. 2024Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Controlcited by 22 here
  4. 2024TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policiescited by 25 here
  5. 2025VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Cachingcited by 38 here
  6. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 110 here
  7. 2025HybridVLA: Collaborative Diffusion and Autoregression in a Unified Vision-Language-Action Modelcited by 23 here
  8. 2025MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulationcited by 31 here
  9. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Modelscited by 39 here
  10. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 133 here
  11. 2025UniVLA: Learning to Act Anywhere with Task-centric Latent Actionscited by 34 here
  12. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Roboticscited by 88 here
  13. 2025VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Modelcited by 24 here
  14. 2025X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Modelcited by 30 here

The 24 most cited of 437.

Browse

30 of 237, ranked 6–35

  1. 6SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruning202524
  2. 7HybridVLA: Collaborative Diffusion and Autoregression in a Unified Vision-Language-Action Model202523
  3. 8Quantization-Aware Imitation-Learning for Resource-Efficient Robotic Control202422
  4. 9SP-VLA: A Joint Model Scheduling and Token Pruning Approach for VLA Model Acceleration202521
  5. 10VLASH: Real-Time VLAs via Future-State-Aware Asynchronous Inference202521
  6. 11Leave No Observation Behind: Real-time Correction for VLA Action Chunks202520
  7. 12Spec-VLA: Speculative Decoding for Vision-Language-Action Models with Relaxed Acceptance202519
  8. 13Discrete Diffusion VLA: Bringing Discrete Diffusion to Action Decoding in Vision-Language-Action Policies202519
  9. 14The Better You Learn, The Smarter You Prune: Towards Efficient Vision-language-action Models via Differentiable Token Pruning202517
  10. 15Training-Time Action Conditioning for Efficient Real-Time Chunking202515
All 30
  1. 16CogVLA: Cognition-Aligned Vision-Language-Action Model via Instruction-Driven Routing & Sparsification202514
  2. 17BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation202513
  3. 18CEED-VLA: Consistency Vision-Language-Action Model with Early-Exit Decoding202513
  4. 19VQ-VLA: Improving Vision-Language-Action Models via Scaling Vector-Quantized Action Tokenizers202513
  5. 20Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models202512
  6. 21FASTER: Rethinking Real-Time Flow VLAs202611
  7. 22Dita: Scaling Diffusion Transformer for Generalist Vision-Language-Action Policy202511
  8. 23How Fast Can I Run My VLA? Demystifying VLA Inference Performance with VLA-Perf202610
  9. 24Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation202510
  10. 25QVLA: Not All Channels Are Equal in Vision-Language-Action Model's Quantization20269
  11. 26Unified Diffusion VLA: Vision-Language-Action Model via Joint Discrete Denoising Diffusion Process20259
  12. 27Learning Native Continuation for Action Chunking Flow Policies20268
  13. 28dVLA: Diffusion Vision-Language-Action Model with Multimodal Chain-of-Thought20258
  14. 29FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via Neural Action Tokenization20258
  15. 30RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models20257
  16. 31EdgeVLA: Efficient Vision-Language-Action Models20257
  17. 32SQAP-VLA: A Synergistic Quantization-Aware Pruning Framework for High-Performance Vision-Language-Action Models20257
  18. 33Verifier-free Test-Time Sampling for Vision-Language-Action Models20257
  19. 34QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models20266
  20. 35KERV: Kinematic-Rectified Speculative Decoding for Embodied VLA Models20266