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

TL;DR

This paper introduces QVLA, a novel action-centric quantization framework for Vision-Language-Action models that significantly reduces model size and computational requirements while preserving high performance, enabling deployment on resource-limited robotic platforms.

Contribution

QVLA presents a channel-wise, importance-guided quantization method specifically designed for embodied control, improving over uniform-bit approaches by considering action-space sensitivity.

Findings

Achieves 29.2% VRAM usage with 98.9% performance retention.

Outperforms LLM-based quantization methods by 22.6%.

Provides a unified framework for quantization and pruning in VLA models.

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.