Back to YOUIBOT
Xunxiao

Xunxiao

巡霄

Last updated: May 31, 2026
Locomotion Type
Wheeled
Stage
Demo
Height
100cm
Release Year
2024

Product Overview

“Xunxiao” is designed for large-scale complex indoor environments and features long endurance and high flexibility. Based on YOUIBOT’s accumulated expertise in semiconductor and energy scenarios, it has already been applied in semiconductor manufacturing Sub-FAB operations and power distribution room operations in the energy industry. Leveraging its scenario adaptability, the YOUIBOT–Xi’an Jiaotong University Embodied Intelligence Robotics Institute team has built an embodied intelligence foundation model with a “one brain, multiple embodiments” architecture. It adopts a hybrid structure combining a multimodal general-purpose foundation model and a “one brain, multiple embodiments” edge-side embodied model, and has completed preliminary validation in real-world applications. Specifically, the foundation model uses a VLM (Vision-Language Model) and MoE (Mixture of Experts) architecture. Based on a multimodal pretraining dataset and a massive expert-level real-world training dataset called “Data Ocean,” it achieves scenario perception for complex tasks and general language understanding, and routes complex instructions to the edge-side embodied model to enable high-frequency real-time control and online quality assessment.

Key Highlights

“Xunxiao” is designed for large-scale complex indoor environments and features long endurance and high flexibility. Based on YOUIBOT’s accumulated expertise in semiconductor and energy scenarios, it has already been applied in semiconductor manufacturing Sub-FAB operations and power distribution room operations in the energy industry.
Leveraging its scenario adaptability, the YOUIBOT–Xi’an Jiaotong University Embodied Intelligence Robotics Institute team has built an embodied intelligence foundation model with a “one brain, multiple embodiments” architecture. It adopts a hybrid structure combining a multimodal general-purpose foundation model and a “one brain, multiple embodiments” edge-side embodied model, and has completed preliminary validation in real-world applications.
Specifically, the foundation model uses a VLM (Vision-Language Model) and MoE (Mixture of Experts) architecture. Based on a multimodal pretraining dataset and a massive expert-level real-world training dataset called “Data Ocean,” it achieves scenario perception for complex tasks and general language understanding, and routes complex instructions to the edge-side embodied model to enable high-frequency real-time control and online quality assessment.

Target Scenarios

Outdoor patrolinfrastructure inspection
Subscribe to Weekly Report

Get weekly updates on humanoid robot funding, new companies, and product launches.