Search papers, labs, and topics across Lattice.
This paper introduces VQVLA, a motion-aware vector quantization framework designed to enhance the efficiency of Vision-Language-Action (VLA) model inference by leveraging weight similarity and execution dynamics. By implementing a motion-aware quantization scheme, the framework dynamically adjusts precision based on the robot's state, significantly reducing memory access while maintaining task performance. Experimental results demonstrate that VQVLA achieves substantial speedups over existing accelerators, with up to 6.5x improvement compared to the A100 GPU, all while preserving accuracy.
VQVLA accelerates VLA inference by up to 6.5x through innovative motion-aware quantization and centroid reuse, all without sacrificing accuracy.
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.
