视觉-语言-动作模型压缩

本项目主要关注视觉-语言-动作（Vision-Language-Action, VLA）模型的压缩方法, 包括:

• 综述
• 视觉输入压缩
• 骨干网络压缩
• 动作生成压缩
• 混合策略压缩

更新时间: 2026年08月07日

综述

• [2026.06] 面向资源受限具身智能的视觉-语言-动作（VLA）模型综述, 自动化学报[Paper]
• [2026.04] Vision-Language-Action in Robotics: A Survey of Datasets, Benchmarks, and Data Engines, TMLR [Paper]
• [2025.11] 面向具身操作的视觉-语言-动作模型综述, 自动化学报 [Paper]
• [2025.10] Efficient Vision-Language-Action Models for Embodied Manipulation: A Systematic Survey, arXiv [Paper]
• [2025.10] A Survey on Efficient Vision-Language-Action Models, arXiv [Paper]
• [2025.09] 视觉-语言-动作模型综述：从前史到前沿, 自动化学报 [Paper]
• [2025.09] Vision-Language-Action Models for Robotics: A Review Towards Real-World Applications, IEEE Access[Paper]

视觉输入压缩

视觉词元剪枝

• [2026.03] VLA-IAP: Training-Free Visual Token Pruning via Interaction Alignment for Vision-Language-Action Models, arXiv [Paper]
• [2025.11] VLA-Pruner: Temporal-Aware Dual-Level Visual Token Pruning for Efficient Vision-Language-Action Inference, arXiv [Paper] [Code]
• [2025.09] Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation, arXiv [Paper]
• [2025.09] SpecPrune-VLA: Accelerating Vision-Language-Action Models via Action-Aware Self-Speculative Pruning, arXiv [Paper]
• [2025.08] CogVLA: Cognition-Aligned Vision-Language-Action Model via Instruction-Driven Routing & Sparsification, NeurIPS 2025 [Paper]

视觉词元合并

• [2026.07] NativeMEM: Native Memory Compression for Long-Horizon Robotic Manipulation, arXiv [Paper]
• [2026.06] Fast Enough to Act: Spatio-Temporal Visual Token Merging for Low-Latency Robotic VLMs and VLAs, arXiv [Paper]
• [2026.03] SemanticVLA: Semantic-Aligned Sparsification and Enhancement for Efficient Robotic Manipulation, AAAI [Paper][Code]
• [2026.03] DepthCache: Depth-Guided Training-Free Visual Token Merging for Vision-Language-Action Model Inference, arXiv [Paper]
• [2025.12] Token Expand-Merge: Training-Free Token Compression for Vision-Language-Action Models, RA-L [Paper]
• [2025.11] COMPRESSOR-VLA: Instruction-Guided Visual Token Compression for Efficient Robotic Manipulation, arXiv [Paper]
• [2025.09] Focusing on What Matters: Object-Agent-centric Tokenization for Vision Language Action Models, CoRL [paper]

视觉词元缓存

• [2026.03] TTF-VLA: Temporal Token Fusion via Pixel-Attention Integration for Vision-Language-Action Models, AAAI [Paper] [Code]
• [2026.02] Learning to Accelerate Vision-Language-Action Models through Adaptive Visual Token Caching, arXiv [Paper]
• [2025.09] RetoVLA: Reusing Register Tokens for Spatial Reasoning in Vision-Language-Action Models, ICRA 2026 [Paper] [Code]
• [2025.02] VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching, NeurIPS 2025 [Paper] [Code]

骨干网络压缩

权重压缩

• [2026.07] A Motion-Aware Vector Quantization Framework with Centroid Reuse for Efficient VLA Inference, arXiv [Paper]
• [2026.03] DyQ-VLA: Temporal-Dynamic-Aware Quantization for Embodied Vision-Language-Action Models, arXiv [Paper]
• [2025.10] Don't Run with Scissors: Pruning Breaks VLA Models but They Can Be Recovered, arXiv [Paper]
• [2025.06] RLRC: Reinforcement Learning-based Recovery for Compressed Vision-Language-Action Models, RA-L [Paper]

层级压缩

• [2026.06] Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?, arXiv [Paper][Code]
• [2026.06] BLUE: Toward Better Language Use in Efficient Vision-Language-Action Models for Autonomous Driving, arXiv [Paper][Code]
• [2026.02] DySL-VLA: Efficient Vision-Language-Action Model Inference via Dynamic-Static Layer-Skipping for Robot Manipulation, arXiv [Paper]
• [2025.11] ActDistill: General Action-Guided Self-Derived Distillation for Efficient Vision-Language-Action Models, arXiv [Paper]
• [2025.03] MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation, AAAI [Paper][Code]
• [2024.11] DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution, NeurIPS [Paper] [Code]

缓存压缩

• [2025.09] KV-Efficient VLA: A Method to Speed up Vision Language Models with RNN-Gated Chunked KV Cache, arXiv [Paper]

动作生成压缩

• [2026.07] ActionCache: Training-Free Acceleration for Vision-Language-Action Models with Action Caching and Refinement, arXiv [Paper]
• [2026.06] NAC: Neural Action Codec for Vision-Language-Action Models, arXiv [Paper]
• [2026.06] ELASTIC: Efficiently Learning to Adaptively Scale Test-Time Compute for Generative Control Policies, arXiv [Paper]
• [2026.06] TBD-VLA: Temporal Block Diffusion Vision Language Action Model, arXiv [Paper][Code]
• [2025.06] CEED-VLA: Consistency Vision-Language-Action Model with Early-Exit Decoding, arXiv [Paper][Code]
• [2025.06] Fast ECoT: Efficient Embodied Chain-of-Thought via Thoughts Reuse, arXiv [Paper]
• [2025.01] FAST: Efficient Action Tokenization for Vision-Language-Action Models, arXiv [Paper]

混合策略压缩

联合压缩

• [2026.07] Reflex: Real-Time VLA Control through Streaming Inference, arXiv [Paper]
• [2026.06] Mix-QVLA: Task-Evidence-Aware Mixed-Precision Quantization of Vision-Language-Action Models, arXiv [Paper]
• [2026.05] Ω-QVLA: Robust Quantization for Vision-Language-Action Models via Composite Rotation and Per-step Scaling, arXiv [Paper][Code]
• [2026.04] DA-PTQ: Drift-Aware Post-Training Quantization for Efficient Vision-Language-Action Models, arXiv [Paper]
• [2026.02] QVLA: Not All Channels Are Equal in Vision-Language-Action Model's Quantization, ICLR 2026 [Paper] [Code]
• [2026.02] HBVLA: Pushing 1-Bit Post-Training Quantization for Vision-Language-Action Models, arXiv [Paper]
• [2026.02] QuantVLA: Scale-calibrated post-training quantization for vision-language-action models, CVPR 2026 [Paper] [Project]
• [2025.09] SQAP-VLA: A Synergistic Quantization-Aware Pruning Framework for High-Performance Vision-Language-Action Models, arXiv [Paper][Code]
• [2025.06] EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models, NeurIPS 2025 [Paper] [Code]

动态调度

• [2026.04] A1: A Fully Transparent Open-Source, Adaptive and Efficient Truncated Vision-Language-Action Model, arXiv [Paper][Code]
• [2026.02] EcoVLA: Environment-Aware Adaptive Pruning with Interleaved Inference Orchestration for Vision-Language-Action Models, arXiv [Paper]
• [2026.01] AC²-VLA: Action-Context-Aware Adaptive Computation in Vision-Language-Action Models for Efficient Robotic Manipulation, arXiv [Paper]
• [2025.08] Leveraging OS-Level Primitives for Robotic Action Management, arXiv [Paper]
• [2025.06] SP-VLA: A Joint Model Scheduling and Token Pruning Approach for VLA Model Acceleration, arXiv [Paper]
• [2025.05] Think Twice, Act Once: Token-Aware Compression and Action Reuse for Efficient Inference in Vision-Language-Action Models, arXiv [Paper]

轻量架构

• [2026.07] FibVLA: An Efficient Temporal Vision-Language-Action Model with Fibonacci Sampling, arXiv [Paper]
• [2026.07] FabriVLA: A Lightweight Vision-Language-Action Model for Precise Multi-Task Manipulation, arXiv [Paper]
• [2026.07] Teaching Tiny VLA Models Where to Look and How to Move, arXiv [Paper]
• [2026.06] ReactVLA: Fast and Lightweight Reactive Robot Manipulation via Improved Mean Flow Action Generation, arXiv [Paper]
• [2026.06] SpikeVLA: Vision-Language-Action Models with Spiking Neural Networks, ICML [Paper]
• [2026.01] Shallow-π: Knowledge Distillation for Flow-based VLAs, arXiv [Paper]
• [2025.09] FLOWER: Democratizing Generalist Robot Policies with Efficient Vision-Language-Action Flow Policies, arXiv [Paper]
• [2025.09] dVLA: Diffusion Vision-Language-Action Model with Multimodal Chain-of-Thought, arXiv [Paper]
• [2025.06] SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics, arXiv [Paper][Code]
• [2025.06] BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation, arXiv [Paper][Code]

致谢

感谢徐力同学协助整理