AGQ: Action-Guided Ultra-Low-Bit Quantization for Vision-Language-Action Models

Xuhong Qiu; Zhuoyu Wu; Qiawei Zheng; Wenhui Ou; Wenqi Fang; Raphaël C. -W. Phan

cs.CVOct 21, 2026

AGQ: Action-Guided Ultra-Low-Bit Quantization for Vision-Language-Action Models

Authors: Xuhong Qiu, Zhuoyu Wu, Qiawei Zheng, Wenhui Ou, Wenqi Fang, Raphaël C. -W. Phan

Organizations: School of Computer Science and Engineering, Southeast University, Nanjing, China · School of Information Technology, Monash University, Subang Jaya, Malaysia · Department of Electronic & Computer Engineering, Hong Kong University of Science & Technology, Hong Kong, China · Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Abstract

Vision-language-action (VLA) models enable robots to perform tasks specified through natural-language instructions, but their large weight footprint challenges deployment on memory-constrained onboard platforms. Ultra-low-bit quantization reduces this footprint, yet preserving task performance and controlling weight recovery costs remain challenging. We propose Action-Guided Quantization, which combines action sensitivity with a measured proxy for weight recovery cost to allocate matrix bitwidths under storage budgets. Per-channel calibration compensates for quantization error, while a uniform bitwidth within each matrix preserves regular layouts. Bit-dependent grouped lossless coding reduces resident weight storage, and block-wise recovery limits temporary buffers. On LIBERO, AGQ achieves 92.7% task success at W3, exceeding AWQ and GPTQ by 1.2 and 3.7 percentage points, respectively, and retains 70.5% success at W2.5. An asymmetric configuration, denoted W1.5 $^{*}$ , uses W4 vision weights and a parameter-weighted average of 1.5 bits for language weights. With lossless coding, it achieves $6.10\times$ complete-model compression relative to BF16 while retaining 25.5% task success. Code and experimental results are available at https://github.com/Qiuxu00/AGQ

Figures & tables

Fig. 1: Low-bit quantization of VLA models. Lines show task success rates (left axis) for Ours and Baseline (QVLA), while bars show target-weight compression ratios relative to BF16 (right axis). QVLA reports 0% at W2.5; lower-bit results are zero-valued extensions for visualization. W1.5*: W4 vision + W1.5 language.

Fig. 2: Overview of AGQ: (a) VLA architecture; (b) action-guided bit allocation incorporating a measured proxy for weight recovery cost; (c) per-channel quantizer calibration; and (d) grouped lossless coding and block-wise execution.

Budget	$\boldsymbol{\lambda}$	Total (428)
Budget	$\boldsymbol{\lambda}$	W1	W2	W3	W4
W2.5	0	54 (23+31)	107 (67+40)	189 (63+126)	78 (51+27)
	0.25	28 (8+20)	142 (91+51)	192 (55+137)	66 (50+16)
	0.5	19 (6+13)	150 (94+56)	199 (51+148)	60 (53+7)
	0.75	14 (5+9)	155 (96+59)	199 (45+154)	60 (58+2)
	1	7 (5+2)	167 (95+72)	196 (47+149)	58 (57+1)

TABLE I: Matrix-wise bit allocation with supporting probe analysis

Fig. 3: W2.5 success rates across four suites and overall. Random allocation also undergoes calibration. QVLA values are from ActQuant [ 7 ] .

Model	Method	Vision+ LLM BPW	Success Rate (%)
Model	Method	Vision+ LLM BPW	Spatial	Object	Goal	Long	Avg.	$\Delta$
	Baseline	16.0	97.6	98.4	96.8	95.1	96.9	0.0
		4.0	94.6	98.6	96.8	94.4	96.1	$-0.8$
		3.0	86.4	98.0	90.2	91.4	91.5	$-5.4$
	AWQ	2.0	0.0	0.0	0.0	0.0	0.0	$-96.9$
		4.0	90.8	98.8	96.4	92.6	94.6	$-2.3$

TABLE II: Closed-loop task success rates (%) on LIBERO

Model	Method	Vision+ LLM BPW	Model Size (GB)	Peak GPU Mem. (GB)	Latency (ms)
	BF16	16.0	15.457 (1.00 $\times$ )	16.082	187.3
		4.0	4.680 (3.30 $\times$ )	5.643	249.3
		3.0	3.774 (4.10 $\times$ )	4.727	254.4
		2.5	3.327 (4.65 $\times$ )	4.280	245.2
		2.0	2.878 (5.37 $\times$ )	3.828	243.6
	Ours	1.5 ∗	2.656 (5.82 $\times$ )	3.603	250.5

TABLE III: Storage and inference efficiency of OpenVLA-OFT

VLM Quantization

Vision-language model quantization reduces model precision to lower memory and computation while preserving multimodal capabilities.

See all 52 papers

ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models
May 19, 2026Arash Akbari, Arman Akbari, Masih Eskandar +11Neural Network Quantization VLM Quantization
1Northeastern University · University of Georgia · 3Cisco Systems
FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding
Sep 21, 2026Hung T. Ho, Khanh D. Nguyen, Quang D. Nguyen +5VLM Quantization Efficient VLA Model Inference
VinRobotics, Vietnam · AICV Lab, EECS Department, University of Arkansas · School of Advanced Manufacturing and Robotics, Peking University, China +2
QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models
Feb 23, 2026Jingxuan Zhang, Yunta Hsieh, Zhongwei Wan +5Efficient VLA Model Inference VLA Model Post-Training
The Ohio State University · University of Michigan · City University of Hong Kong

May 19, 2026J/K move · Enter open · S save

ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models

Arash Akbari, Arman Akbari, Masih Eskandar, Qitao Tan+10

1Northeastern University · University of Georgia · 3Cisco Systems

Neural Network Quantization VLM Quantization Efficient VLA Model Inference Low-Bit Quantization Mixed-Precision Quantization Efficient VLA Models+4

Open paper AskPDF arXiv

Vision-Language-Action (VLA) models exhibit remarkable action generation for embodied intelligence, but their heavy compute make deployment on edge platforms impractical. Aggressive, sub-4-bit weight quantization is the natural solution, yet existing post-training quantization (PTQ) methods suffer severe performance degradation in this regime. To address this, we introduce ActQuant, an action-guided mixed-precision PTQ framework that operates in two stages: (1) an inter-tensor bit allocator that assigns each weight matrix a single bit-width based on how much it contributes to predicting the agent's actions; (2) an intra-tensor scale optimizer tunes per-block quantization scales using action-aware curvature, so that dynamic range is concentrated on the weights most influential for control. To deliver the on-device benefits of our aggressive quantization, we further introduce OmniModel.cpp, an agentic conversion pipeline that ports architectures into a native C/C++ runtime with efficient low-bit kernels. We evaluate ActQuant both in simulation and on a real-world 6-DoF UR3 arm, with all models deployed through OmniModel.cpp. On the LIBERO benchmark, ActQuant is the only method that operates at or below 3 bits-per-weight, retaining 95.0% on OpenVLA-OFT and 94.8% on

π_{0.5}

. Pushed further, ActQuant reaches 2.5 bpw at 90.1% on OpenVLA-OFT, compressing the backbone from 14.3 GB to 2.7 GB (5.3

\times

). On the physical UR3 arm,

π_{0.5}

quantized with ActQuant retains the baseline's success rate while reducing the memory footprint by 2.5

\times

Abstract

VLM Quantization

Explore similar work

ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models

FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent Folding

QuantVLA: Scale-Calibrated Post-Training Quantization for Vision-Language-Action Models