Gasgoo Munich- PokeBot, a general embodied intelligence company, has closed a nine-figure Pre-A funding round, according to Gasgoo Embodied Intelligence.
The round was co-led by Shunwei Capital and Matrix Partners, with participation from financial and industrial investors including Ubiquant Ventures, Junshan Capital, SEE Fund, Liepin Investment, Yuannuo Capital, and ETC Capital. Existing backers, including Yunqi Partners, Xiaomi Strategic Investment, Highlight Capital, InnoAngel Fund, and Oriental Fortune, doubled down.
Notably, this marks the startup's second funding round in just three months since its founding.
In late April, PokeBot announced an angel round worth tens of millions of dollars, led by Yunqi Partners. That round drew support from top-tier U.S. dollar funds like Shunwei Capital and Highlight Capital, strategic players including Xiaomi Strategic Investment and Xinghaitu, and market-oriented funds like BV Baidu Ventures.
Founded in April 2026, PokeBot focuses on getting robots to work in the physical world. The startup aims to build general-purpose robots that understand physical laws and learn continuously. Its goal is to handle high-dexterity, long-horizon complex tasks in home and service environments—shifting robots from simply "moving" to truly "working."
Image Source: PokeBot
PokeBot has established a full technology stack—spanning foundation models, real-world reinforcement learning, data collection hardware, and complex task training. It has also assembled a team covering foundation models, reinforcement learning, agents, robot hardware, systems, and commercialization.
Technically, PokeBot isn't training separate models for every task. Instead, it's building a general embodied intelligence system that achieves cross-task capability transfer by understanding physical causality.
To do this, PokeBot has built a technical flywheel combining three elements: WAM (World Action Model), RL (full-stack real-world reinforcement learning), and DATA (high-quality real-world data).
WAM enables the model to grasp the physical cause-and-effect relationship between actions and environmental changes. By tightly coupling the world model's predictive power with robotic decision-making, it can accurately determine how actions will alter the environment.
The full-stack RL system covers training environments, reward models, long-horizon action evaluation, and a unified multi-task architecture. This allows robots to evolve continuously based on successes and failures from real-world interactions.
High-quality real-world data serves as the foundation for the continuous evolution of both WAM and RL. PokeBot has developed proprietary data collection hardware for fine manipulation, lowering the barrier to gathering high-quality data and closing the loop between data collection, foundation model training, real-world reinforcement learning, and new data feedback.
In June, PokeBot released a demo of a robot autonomously cooking Mapo Tofu. The roughly 9-minute task marks the world's first complete, long-horizon Chinese cooking operation performed by a physical robot. It tackled five core challenges in robotics: error control across long task sequences, fine force control on soft objects like tofu, real-time adjustments in dynamic cooking environments, multi-tool coordination with autonomous error correction, and millimeter-level precision alignment.
Beyond Mapo Tofu, PokeBot demonstrated other high-difficulty operations, including folding clothes, threading zip ties, and tying sachets. These scenarios cover soft object manipulation, fine assembly, and long-horizon tasks.
About Seeds Discovery:
Gasgoo's "Seeds Discovery" column aims to build a service platform connecting startups, ecosystem partners, investors, and local governments to empower the automotive supply chain. Since its launch, the column has sought to identify promising companies, technologies, and business models driving the intelligent transformation of the industry. According to Gasgoo, nearly every startup featured in "Seeds Discovery" has successfully connected with resources across the automotive ecosystem.
