mlsanityData to 7 Oct 2026

VLAQuantBench: Closed-Loop Evaluation of Post-Training Quantization for Vision-Language-Action Models

arXiv 2609.25376 ↗First posted 21 Sep 2026cs.RO cs.AIcited by 0

On the map: Vision-language-action models → VLA architectures and generalization

Abstract (arXiv)

Post-training quantization reduces the memory requirements of vision-language-action (VLA) models, but precision selection must account for the interaction between layer scope, numerical format, and calibration. We introduce VLAQuantBench, a controlled evaluation with 409 runs and 94,574 simulation episodes: four models on LIBERO, with X-VLA additionally evaluated on three simulation benchmark families. Under uncalibrated W4A4 round-to-nearest quantization, expanding a π0.5 action-head subset from 126 to 167 layers raises success from 7.0% to 70.5%. Fixed-observation replay confirms a corresponding numerical recovery. Two-episode calibration removes the severe joint failures in the tested subsets, whereas the same smoothing-and-clipping recipe lowers π0 success and does not recover OpenVLA-OFT end-to-end. For OpenVLA-OFT, protecting one 28,672-parameter output projection instead restores near-baseline success: the remaining 441 eligible linear layers retain W3 on LIBERO-Long or eight-bit activations across all four suites. Task-clustered intervals support the large failure and recovery contrasts. These results establish recipe-dependent interactions and identify concrete precision assignments, rather than universal layer-sensitivity rules. Real-kernel and physical-robot measurements complement the accuracy analysis. Code, configurations, and episode records are publicly available at https://github.com/jiuyixu25/VLAQuantBench.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in VLA architectures and generalization, one of 13 directions in Vision-language-action models. Builds on 34 of its 36 references that are in the map. First posted on arXiv 21 Sep 2026.

Builds on

34 of 36 references.

References · 34

  1. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,355
  2. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  4. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  5. 2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,007
  6. 2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,838
  7. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  8. 2017Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inferencecited by 1,548
  9. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  10. 2022SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Modelscited by 1,338
  11. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  12. 2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,234
22 more
  1. 2022LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scalecited by 881
  2. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 844
  3. 2024DROID: A Large-Scale In-The-Wild Robot Manipulation Datasetcited by 801
  4. 2020Up or Down? Adaptive Rounding for Post-Training Quantizationcited by 600
  5. 2021CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Taskscited by 564
  6. 2018Quantizing deep convolutional networks for efficient inference: A whitepapercited by 507
  7. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Roboticscited by 507
  8. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  9. 2021BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstructioncited by 464
  10. 2019HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precisioncited by 387
  11. 2024TinyVLA: Towards Fast, Data-Efficient Vision-Language-Action Models for Robotic Manipulationcited by 331
  12. 2019Data-Free Quantization Through Weight Equalization and Bias Correctioncited by 328
  13. 2025X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Modelcited by 285
  14. 2022QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantizationcited by 175
  15. 2024VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning Taskscited by 171
  16. 2025BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulationcited by 42
  17. 2026QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Modelscited by 25
  18. 2026QVLA: Not All Channels Are Equal in Vision-Language-Action Model's Quantizationcited by 23
  19. 2025SQAP-VLA: A Synergistic Quantization-Aware Pruning Framework for High-Performance Vision-Language-Action Modelscited by 15
  20. 2026DyQ-VLA: Temporal-Dynamic-Aware Quantization for Embodied Vision-Language-Action Modelscited by 10
  21. 2023Mixed-Precision Quantization for Deep Vision Models with Integer Quadratic Programmingcited by 6
  22. 2026HBVLA: Pushing 1-Bit Post-Training Quantization for Vision-Language-Action Modelscited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026HoloQ-VLA: Uniform W4A4 Quantization of Vision-Language-Action Models
  2. 2026CHASE-VLA: Post-Training Quantization Framework for Vision-Language-Action Models with Chunk-Aware Scale Estimation
  3. 2026ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models
  4. 2025EaqVLA: Encoding-aligned Quantization for Vision-Language-Action Models
  5. 2025EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models
  6. 2024Q-VLM: Post-training Quantization for Large Vision-Language Models
  7. 2026FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding
  8. 2026VLA Foundry: A Unified Framework for Training Vision-Language-Action Models