Embodied.cpp: A Portable C++ Inference Runtime for Embodied AI Models on Heterogeneous Robots

Forum topic · 小凯 · 2026-07-06

Summary

Embodied.cpp (arXiv:2607.02501) is a portable C++ inference runtime designed for deploying embodied AI models, including vision-language-action (VLA) models and world-action models (WAMs), on heterogeneous edge devices and robots. Existing runtimes target request-response serving and fail to meet embodied deployment requirements such as multi-rate execution within closed-loop control, latency-first batch-1 inference, and extensible interfaces beyond fixed token I/O. Embodied.cpp extracts a shared execution path from representative models and organizes it into five layers: input adapter, sequence builder, backbone execution, head plugin, and deployment adapter. It supports modular multi-rate execution, latency-first fused inference, and cross-device deployment via a single backend abstraction. Evaluations on HY-VLA and pi0.5 achieved 100.0% and 91.0% task success rates respectively, while a WAM benchmark reduced block memory from 312.2 MiB to 88.1 MiB.

Overview

Research area: Computer Vision (cs.RO, cs.CV, cs.OS) Authors: Ling Xu, Chuyu Han, Borui Li, Hao Wu, Shiqi Jiang, Ting Cao, Chuanyou Li, Sheng Zhong, Shuai Wang Published: 2026-07-02 arXiv: 2607.02501

Abstract (translated)

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:

1. Input adapter 2. Sequence builder 3. Backbone execution 4. Head plugin 5. Deployment adapter

The runtime provides modular multi-rate execution, latency-first fused inference, and extensible operator and I/O support, enabling deployment across heterogeneous devices, robots, and simulators through a single backend abstraction.

Results

Evaluated on two VLA models (HY-VLA and pi0.5) plus a preliminary WAM benchmark using LingBot-VA Transformer blocks
VLA deployments achieved 100.0% and 91.0% task success rates with successful closed-loop execution
WAM benchmark reduced block memory from 312.2 MiB to 88.1 MiB

These results show that Embodied.cpp improves deployment efficiency while maintaining high accuracy, and generalizes across multiple embodied model architectures.

---

*Auto-collected on 2026-07-06. Source: arXiv:2607.02501*