QVLA: Not All Channels Are Equal in Vision-Language-Action Model's Quantization

2026-02-03

AI Summary

Existing vision-language-action (VLA) models are challenging to deploy on resource-constrained robotic platforms due to their high computational overhead, and conventional uniform quantization neglects the action space’s high sensitivity to minor errors, often leading to task failure. This work proposes QVLA, the first action-centric quantization framework for embodied control, which introduces a channel-wise action sensitivity analysis to derive importance metrics and jointly optimizes non-uniform bit-width allocation and pruning within a unified global optimization framework. Evaluated on the LIBERO benchmark, QVLA reduces the memory footprint of OpenVLA-OFT by 70.8% (to 29.2% of the original), retains 98.9% of its performance, achieves a 1.49× speedup, and outperforms SmoothQuant by 22.6% in task success rate.

Technology Category

• Computer Vision: Large Vision Models
• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

The advent of Vision-Language-Action (VLA) models represents a significant leap for embodied intelligence, yet their immense computational demands critically hinder deployment on resource-constrained robotic platforms. Intuitively, low-bit quantization is a prevalent and preferred technique for large-scale model compression. However, we find that a systematic analysis of VLA model's quantization is fundamentally lacking. We argue that naively applying uniform-bit quantization from Large Language Models (LLMs) to robotics is flawed, as these methods prioritize passive data fidelity while ignoring how minor action deviations compound into catastrophic task failures. To bridge this gap, we introduce QVLA, the first action-centric quantization framework specifically designed for embodied control. In a sharp departure from the rigid, uniform-bit quantization of LLM-based methods, QVLA introduces a highly granular, channel-wise bit allocation strategy. Its core mechanism is to directly measure the final action-space sensitivity when quantizing each individual channel to various bit-widths. This process yields a precise, per-channel importance metric that guides a global optimization, which elegantly unifies quantization and pruning (0-bit) into a single, cohesive framework. Extensive evaluations on different baselines demonstrate the superiority of our approach. In the LIBERO, the quantization version of OpenVLA-OFT with our method requires only 29.2% of the original model's VRAM while maintaining 98.9% of its original performance and achieving a 1.49x speedup. This translates to a 22.6% performance improvement over the LLM-derived method SmoothQuant. Our work establishes a new, principled foundation for compressing VLA models in robotics, paving the way for deploying powerful, large-scale models on real-world hardware. Code will be released.

Problem

• Vision-Language-Action
• quantization
• embodied intelligence
• action sensitivity
• model compression

Innovation

• Vision-Language-Action
• channel-wise quantization
• action-centric compression
• bit allocation
• quantization-pruning co-optimization