Gasgoo Munich-Xieyue Intelligence has secured hundreds of millions of yuan in an Angel+ funding round. The round was backed by investors including Linear Capital, PEAKVEST, Rhein Capital, and Hidden Hill Capital.
The capital will fuel the company's push to train embodied foundation models, build computing and data infrastructure, and expand its core team. Funds also support the development and scenario verification of home robot hardware.
This marks the startup's second funding round within just seven months. A prior Angel round was led by Li Auto and Vision Plus Capital.
The rapid fundraising is backed by a dual advantage: a top-tier team and a differentiated market strategy.
Image source: Xieyue Intelligence
Xieyue Intelligence was co-founded by Chen Wei, former chief scientist of AI and head of the foundation model department at Li Auto, alongside Zhang Xiao, the automaker's former product line president. The core team is drawn from leading global tech, AI, robotics, and intelligent automotive firms, spanning five key areas: large model algorithms, AI infrastructure, robot motion control, and product and supply chain management. It is one of the few domestic teams with experience in training large models across 10,000 GPU cards, developing digital-physical dual models, and managing the full mass-production chain for consumer products.
Strategically, Xieyue Intelligence is diverging from the industry's rush into industrial and logistics scenarios. Instead, it is positioning the home as the core training ground, testing bed, and scaling environment for its embodied foundation models. The company views the household as one of the most complex, high-frequency, and long-term valuable physical environments—making it an ideal touchstone for validating general embodied intelligence.
That said, Xieyue is adopting a gradual "verify first, enter home later" strategy. It plans to prioritize semi-structured environments like hotels and nursing homes to prove model capabilities and unit economics before gradually moving into household settings. Initial targets include high-frequency chores like laundry, organizing, and cleaning.
To achieve this, Xieyue is building a closed-loop ecosystem encompassing robot hardware, embodied foundation models, and home-based self-evolution. The system is driven by "Duplex Reasoning" for the foundation model, a "Human-centric" approach for the physical world data flywheel, and a "Safety-first" framework for robot security. It also features a "Home-native" dual-form body designed to blend naturally into real home environments.
Image source: Xieyue Intelligence
Specifically, Duplex Reasoning breaks away from the traditional simplex or half-duplex workflow of "receive command—execute action—return result." It aims to maintain a "two-way channel" between the robot and human during perception, reasoning, and execution. This allows conversational information to be interrupted or corrected at any time, and enables action goals to be adjusted dynamically during the process.
Technically, Duplex Reasoning uses a Vision-Language-Action (VLA) model as its backbone, integrated with world model predictions. Language compresses visual, audio, and environmental changes into transferable high-level semantics, driving cross-scenario generalization. Meanwhile, the world model predicts potential outcomes of actions, providing pre-constraints for long-horizon tasks.
Both elements target a single goal: boosting the success rate of robot understanding, decision-making, and action in open environments—all while keeping computing costs under control.
At the same time, Xieyue Intelligence argues that the core competitiveness of embodied intelligence lies not in larger models or greater data volume, but in high-quality data and systematic infrastructure capabilities. Training infrastructure and data infrastructure are viewed as two equally critical pillars.
To date, Xieyue has completed the initial development of proprietary "Ego" data collection devices and a data platform. It is gradually building a layered data system centered on first-person perspectives, body-agnostic data, and body-specific high-quality data. The company plans to have the full pipeline—from collection, cleaning, and labeling to training—fully operational by 2026.
On the hardware front, the startup employs a "single body, full-stack closed-loop" strategy. It plans to control hardware variables initially to refine model capabilities before gradually rolling out consumer-facing home robot products.
About Seeds Discovery:
Gasgoo's "Seeds Discovery" column aims to build a service platform connecting startups, industrial ecosystem partners, investment firms, and local governments to deeply empower the automotive supply chain. Since its inception, the column has been dedicated to identifying high-potential companies, technologies, and business models that offer inspiration and leadership amid the wave of intelligent transformation, thereby driving the growth of innovative forces in the auto industry. According to Gasgoo statistics, nearly all startups featured in "Seeds Discovery" have successfully secured connections with industrial ecosystem resources.
