How vision-language-action models scale from power-constrained edge SoCs to datacenter GPUs
Vision-language-action models are becoming the standard policy class for robotic control, but robots ship with edge SoCs while most VLA benchmarking happens on datacenter GPUs.
We evaluated five representative VLA models — state-of-the-art baselines and two newly proposed architectures — on the LIBERO benchmark across edge platforms under varying power budgets and across datacenter GPU configurations, measuring task accuracy jointly with latency, throughput, and peak memory.
Key findings:
Architectural choices (action tokenization, backbone size) strongly influence throughput and memory footprint, independent of raw hardware capability.
Power-constrained edge devices degrade non-linearly — yet well-chosen edge configurations can match or exceed older datacenter GPUs, challenging the assumption that robotic inference belongs in the cloud.
High-throughput VLA variants are achievable without significant accuracy loss.
Published at GLSVLSI 2026 (Taherin et al., 2026); the related VOTE trajectory-ensemble optimization (Lin et al., 2025) reaches 46 Hz throughput on edge platforms with public code.
Vision-Language-Action (VLA) models have emerged as powerful generalist policies for robotic control, yet their performance scaling across model architectures and hardware platforms, as well as their associated power budgets, remain poorly understood. This work presents an evaluation of five representative VLA models—spanning state-of-the-art baselines and two newly proposed architectures—targeting edge and datacenter GPU platforms. Using the LIBERO benchmark, we measure accuracy alongside system-level metrics, including latency, throughput, and peak memory usage, under varying edge power constraints and high-performance datacenter GPU configurations. Our results identify distinct scaling trends: (1) architectural choices, such as action tokenization and model backbone size, strongly influence throughput and memory footprint; (2) power-constrained edge devices exhibit non-linear performance degradation, with some configurations matching or exceeding older datacenter GPUs; and (3) high-throughput variants can be achieved without significant accuracy loss. These findings provide actionable insights when selecting and optimizing VLAs across a range of deployment constraints. Our work challenges current assumptions about the superiority of datacenter hardware for robotic inference.
@inproceedings{taherin2026crossplatform,title={Cross-Platform Scaling of Vision-Language-Action Models from Edge to Cloud GPUs},author={Taherin, Amir and Lin, Juyi and Akbari, Arash and Akbari, Arman and Zhao, Pu and Chen, Weiwei and Kaeli, David and Wang, Yanzhi},booktitle={Proceedings of the Great Lakes Symposium on VLSI (GLSVLSI)},year={2026},doi={10.1145/3787109.3816400},}
Recent large-scale Vision Language Action (VLA) models have shown superior performance in robotic manipulation tasks guided by natural language. However, current VLA models suffer from two drawbacks: (i) generation of massive tokens leading to high inference latency and increased training cost, and (ii) insufficient utilization of generated actions resulting in potential performance loss. To address these issues, we develop a training framework to finetune VLA models for generating significantly fewer action tokens with high parallelism, effectively reducing inference latency and training cost. Furthermore, we introduce an inference optimization technique with a novel voting-based ensemble strategy to combine current and previous action predictions, improving the utilization of generated actions and overall performance. Our results demonstrate that we achieve superior performance compared with state-of-the-art VLA models, achieving significantly higher success rates and 39x faster inference than OpenVLA with 46 Hz throughput on edge platforms, demonstrating practical deployability.
@misc{lin2025vote,title={VOTE: Vision-Language-Action Optimization with Trajectory Ensemble Voting},author={Lin, Juyi and Taherin, Amir and Akbari, Arash and Akbari, Arman and Lu, Lei and Chen, Guangyu and Padir, Taskin and Yang, Xiaomeng and Chen, Weiwei and Li, Yiqian and Lin, Xue and Kaeli, David and Zhao, Pu and Wang, Yanzhi},year={2025},}