[论文] Embodied.cpp: A Portable Inference Runtime of Embodied AI Models on He...

论文概要

研究领域: CV 作者: Ling Xu, Chuyu Han, Borui Li 发布时间: 2026-07-04 arXiv: 2507.03242

中文摘要

具身智能模型现已涵盖视觉-语言-动作（VLA）模型和世界-动作模型（WAMs），但实际部署仍因模型专用的 Python 技术栈、后端假设和机器人端胶水代码而碎片化，尤其在异构边缘设备上。现有推理运行时主要为请求-响应式服务设计，因此无法满足具身部署的运行时契约：闭环控制内的多速率执行、异构硬件上的延迟优先 batch-1 推理，以及超越固定 token I/O 的可扩展具身接口。

本文提出 Embodied.cpp，一个面向具身模型的可移植 C++ 推理运行时。基于对代表性 VLA 模型和 WAM 的架构分析，Embodied.cpp 提取了共享执行路径，并将其组织为五个层次：输入适配器、序列构建器、主干执行、头部插件和部署适配器。该运行时提供模块化多速率执行、延迟优先的融合推理，以及可扩展的算子和 I/O 支持，通过单一后端抽象即可在异构设备、机器人和仿真器上部署。

在两个 VLA 模型 HY-VLA 和 pi0.5 以及基于 LingBot-VA Transformer 块的初步 WAM 基准上的评估显示：VLA 部署分别实现了 100.0% 和 91.0% 的任务成功率；WAM 基准将块内存从 312.2 MiB 降至 88.1 MiB。结果表明 Embodied.cpp 在保持高精度的同时显著提升了部署效率。

原文摘要

Embodied AI models now span vision-language-action (VLA) models and world-action models (WAMs), but practical deployment remains fragmented across model-specific Python stacks, backend assumptions, and robot-side glue code, especially on heterogeneous edge devices. Existing inference runtimes are designed mainly for request-response serving and therefore do not satisfy the runtime contract of embodied deployment: multi-rate execution inside closed-loop control, latency-first batch-1 inference on heterogeneous hardware, and extensible embodied interfaces beyond fixed token I/O. We present Embodied.cpp, a portable C++ inference runtime for embodied models. Based on an architectural analysis of representative VLA models and WAMs, Embodied.cpp captures a shared execution path and organizes it into five layers: input adapters, sequence builders, backbone execution, head plugins, and deployment adapters. The runtime provides modular multi-rate execution, latency-first fused inference, and ex...

--- *自动采集于 2026-07-05*

#论文 #arXiv #CV #小凯