A Motion-Aware Vector Quantization Framework with Centroid Reuse for Efficient VLA Inference
2026-07-27 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors address the problem of slow response times in Vision-Language-Action (VLA) models used for robots by creating a new system called VQVLA. Their method reduces unnecessary memory use and computation by adjusting how data is simplified based on what the robot is doing and by reusing certain calculations. They also build special hardware to support these new techniques. Tests show VQVLA runs much faster than current solutions while keeping accuracy nearly the same.
Vision-Language-Action modelsvector quantizationinference accelerationhardware acceleratorGEMM (General Matrix Multiply)dynamic precisionmotion-aware quantizationcodebook-index representationembodied AIcomputational efficiency
Authors
Zhuoran Song, Haozhe Jiang, Chunyu Qi, Minnan Pei, Gang Li, Xiaoyao Liang, Haibing Guan
Abstract
Vision-Language-Action (VLA) models have demonstrated strong potential for embodied AI, yet their high inference latency on GPUs limits real-time deployment. Existing accelerators, such as Dadu-Corki, improve efficiency but treat VLA models as full-precision workloads, leaving substantial redundancy in both memory and computation underexploited. In this paper, we propose VQVLA, an algorithm-hardware co-design framework that accelerates VLA inference by exploiting weight similarity and execution dynamics. We first introduce MotionVQ, a motion-aware vector quantization scheme that dynamically adjusts quantization precision based on the robot's execution state, reducing memory access while preserving task success rate. We then propose a merged-centroid vectorized GEMM paradigm that operates on the codebook-index representation, eliminating redundant multiplications through spatial aggregation and temporal reuse of centroids. To realize these optimizations, we design an accelerator that efficiently supports dynamic precision selection and centroid-reuse computation. Experimental results show that VQVLA achieves 6.5x, 2.8x, 1.9x, 3.3x, and 4.3x speedup over the A100 GPU, Dadu-Corki, LUT-DLA, CodeGEMM, and ShiftAddLLM, respectively, with negligible accuracy degradation.