HBVLA: Pushing 1-Bit Post-Training Quantization for Vision-Language-Action Models

2026-02-14

AI Summary

This work addresses the challenge of error accumulation in 1-bit quantized vision-language-action (VLA) models, which arises from distributional mismatches in weights and severely degrades long-horizon closed-loop control performance. To mitigate this, the authors propose HBVLA, a binary VLA framework that, for the first time, integrates policy-aware Hessian analysis to identify critical weights, applies sparse orthogonal transformations to non-critical weights, and performs grouped 1-bit quantization in the Haar domain to minimize distributional divergence between full-precision and quantized models. Evaluated on LIBERO and SimplerEnv, HBVLA retains 92.2% and 93.6% of the original performance, respectively, substantially outperforming existing binarization methods. Real-world robot experiments further demonstrate that HBVLA achieves success rates on par with full-precision models, confirming its reliability for deployment on resource-constrained platforms.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Calibration & Uncertainty Quantification
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Vision-Language-Action (VLA) models enable instruction-following embodied control, but their large compute and memory footprints hinder deployment on resource-constrained robots and edge platforms. While reducing weights to 1-bit precision through binarization can greatly improve efficiency, existing methods fail to narrow the distribution gap between binarized and full-precision weights, causing quantization errors to accumulate under long-horizon closed-loop execution and severely degrade actions. To fill this gap, we propose HBVLA, a VLA-tailored binarization framework. First, we use a policy-aware enhanced Hessian to identify weights that are truly critical for action generation. Then, we employ a sparse orthogonal transform for non-salient weights to induce a low-entropy intermediate state. Finally, we quantize both salient and non-salient weights in the Harr domain with group-wise 1-bit quantization. We have evaluated our approach on different VLAs: on LIBERO, quantized OpenVLA-OFT retains 92.2% of full-precision performance; on SimplerEnv, quantized CogAct retains 93.6%, significantly outperforming state-of-the-art binarization methods. We further validate our method on real-world evaluation suite and the results show that HBVLA incurs only marginal success-rate degradation compared to the full-precision model, demonstrating robust deployability under tight hardware constraints. Our work provides a practical foundation for ultra-low-bit quantization of VLAs, enabling more reliable deployment on hardware-limited robotic platforms.

Problem

• Vision-Language-Action models
• 1-bit quantization
• quantization error
• distribution gap
• embodied control

Innovation

• 1-bit quantization
• Vision-Language-Action models
• Hessian-aware binarization
• sparse orthogonal transform
• Harr domain quantization