Deploying embodied AI technology often hits performance snags during edge inference. To tackle this challenge, Wuwen Xinqiong has teamed up with leading academic institutions to co-develop and open-source the APXInf inference engine, tailored specifically for running embodied AI on edge devices. This engine is designed to be compatible with a wide range of computing platforms, substantially cutting down on edge inference latency. It also effectively tackles the issue of inference jitter through a series of multi-layer optimization techniques, proprietary technologies, and a cutting-edge architectural design. Moreover, the engine pioneers the concept of 'agile inference,' which streamlines the process of integrating models. Presently, APXInf supports two types of embodied models and is in the process of broadening its compatibility with additional models and hardware. This open-source project not only bridges a crucial gap in the embodied AI technology ecosystem, establishing a seamless technical loop from model training to deployment on devices, but also makes it easier for developers to implement embodied AI technology. The project team has issued a global call to developers, inviting them to join forces in this collaborative venture.