Characterizing Vision-Language-Action Models across XPUs: Constraints and Acceleration for On-Robot Deployment

Kaijun Zhou, Qiwei Chen, Da Peng, Zhiyang Li, Xijun Li, Jinyu Gu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:164773-164786, 2026.

Abstract

Vision-Language-Action (VLA) models are promising for generalist robot control, but on-robot deployment is bottlenecked by real-time inference under tight cost and energy budgets. Most prior evaluations rely on desktop-grade GPUs, obscuring the trade-offs and opportunities offered by heterogeneous edge accelerators (GPUs/XPUs/NPUs). We present a systematic framework for low-cost VLA deployment via model–hardware co-characterization. First, we build a cross-accelerator leaderboard and evaluate model–hardware pairs under CET (Cost, Energy, Time), showing that “right-sized” edge devices can be more cost-/energy-efficient than flagship GPUs while meeting control-rate constraints. Second, using fine-grained SM tracing and Roofline analysis, we uncover a consistent two-phase inference pattern: a compute-bound VLM backbone followed by a memory-bound Action Expert, which induces phase-dependent underutilization and hardware inefficiency. Finally, guided by these insights, we propose DP-Cache and V-AEFusion to reduce diffusion redundancy and enable asynchronous pipeline parallelism, achieving up to $2.9\times$ speedup on GPUs and $6\times$ on edge NPUs with only marginal success degradation. The example leaderboard website is: https://vla-leaderboard-01.vercel.app/.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-zhou26e, title = {Characterizing Vision-Language-Action Models across {XPU}s: Constraints and Acceleration for On-Robot Deployment}, author = {Zhou, Kaijun and Chen, Qiwei and Peng, Da and Li, Zhiyang and Li, Xijun and Gu, Jinyu}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {164773--164786}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/zhou26e/zhou26e.pdf}, url = {https://proceedings.mlr.press/v306/zhou26e.html}, abstract = {Vision-Language-Action (VLA) models are promising for generalist robot control, but on-robot deployment is bottlenecked by real-time inference under tight cost and energy budgets. Most prior evaluations rely on desktop-grade GPUs, obscuring the trade-offs and opportunities offered by heterogeneous edge accelerators (GPUs/XPUs/NPUs). We present a systematic framework for low-cost VLA deployment via model–hardware co-characterization. First, we build a cross-accelerator leaderboard and evaluate model–hardware pairs under CET (Cost, Energy, Time), showing that “right-sized” edge devices can be more cost-/energy-efficient than flagship GPUs while meeting control-rate constraints. Second, using fine-grained SM tracing and Roofline analysis, we uncover a consistent two-phase inference pattern: a compute-bound VLM backbone followed by a memory-bound Action Expert, which induces phase-dependent underutilization and hardware inefficiency. Finally, guided by these insights, we propose DP-Cache and V-AEFusion to reduce diffusion redundancy and enable asynchronous pipeline parallelism, achieving up to $2.9\times$ speedup on GPUs and $6\times$ on edge NPUs with only marginal success degradation. The example leaderboard website is: https://vla-leaderboard-01.vercel.app/.} }
Endnote
%0 Conference Paper %T Characterizing Vision-Language-Action Models across XPUs: Constraints and Acceleration for On-Robot Deployment %A Kaijun Zhou %A Qiwei Chen %A Da Peng %A Zhiyang Li %A Xijun Li %A Jinyu Gu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-zhou26e %I PMLR %P 164773--164786 %U https://proceedings.mlr.press/v306/zhou26e.html %V 306 %X Vision-Language-Action (VLA) models are promising for generalist robot control, but on-robot deployment is bottlenecked by real-time inference under tight cost and energy budgets. Most prior evaluations rely on desktop-grade GPUs, obscuring the trade-offs and opportunities offered by heterogeneous edge accelerators (GPUs/XPUs/NPUs). We present a systematic framework for low-cost VLA deployment via model–hardware co-characterization. First, we build a cross-accelerator leaderboard and evaluate model–hardware pairs under CET (Cost, Energy, Time), showing that “right-sized” edge devices can be more cost-/energy-efficient than flagship GPUs while meeting control-rate constraints. Second, using fine-grained SM tracing and Roofline analysis, we uncover a consistent two-phase inference pattern: a compute-bound VLM backbone followed by a memory-bound Action Expert, which induces phase-dependent underutilization and hardware inefficiency. Finally, guided by these insights, we propose DP-Cache and V-AEFusion to reduce diffusion redundancy and enable asynchronous pipeline parallelism, achieving up to $2.9\times$ speedup on GPUs and $6\times$ on edge NPUs with only marginal success degradation. The example leaderboard website is: https://vla-leaderboard-01.vercel.app/.
APA
Zhou, K., Chen, Q., Peng, D., Li, Z., Li, X. & Gu, J.. (2026). Characterizing Vision-Language-Action Models across XPUs: Constraints and Acceleration for On-Robot Deployment. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:164773-164786 Available from https://proceedings.mlr.press/v306/zhou26e.html.

Related Material