Galbot's robot is handling a bag of snacks at the exhibition of the 2025 World AI Conference in Shanghai. [Photo provided to chinadaily.com.cn]

As humanoid robots transition from product development to real-world commercialization and deployment, Beijing Galbot Co Ltd is working to crack the code to scalable automation through a radical synthetic data approach.

In an exclusive interview with China Daily, Zhang Zhizheng, co-founder of Galbot, detailed how the company's Sim2Real methodology is overcoming the critical industry barrier, namely, the prohibitively high cost of real-world data collection.

At the heart of Galbot's breakthrough lies in it's pioneering technical paradigm - massive pre-training on large-scale, high-quality synthetic datasets followed by surgical fine-tuning with minimal real-world data. This innovative framework has drastically reduced reliance on real-world data collection.

Zhang said, "We've enabled skills mastered in virtual environments to transfer seamlessly into novel real-world contexts with minimal semantic relabeling." This approach resolves the industry's persistent data scarcity crisis while dramatically enhancing training efficiency and generalization capabilities – establishing the essential groundwork for autonomous, general-purpose robots.

Founded in 2023, Galbot has already raised 2.4 billion yuan ($334.8 million) from big-name investors such as Contemporary Amperex Technology Co Ltd, also known as CATL.

Today, Galbot's humanoid robots are already operating autonomously in over 10 pharmacies in Beijing, performing complex tasks like retrieving medicines from crowded shelves and delivering orders to night-shift couriers – all without human intervention.

"We plan to open 100 such stores nationwide by the end of this year", Zhang said.

Simultaneously, the company's industrial expansion accelerates through its joint venture with a Bosch unit. "Galbot's systems are engineered to meet stringent industrial requirements where failure is not an option," Zhang said, highlighting that "factories demand unwavering accuracy between 99.9 percent and 99.99 percent," necessitating billions of data points to achieve robust generalization.

The company is partnering with Bosch to deploy humanoid robots in factories. The joint venture will focus on applying embodied AI in high-precision manufacturing—such as complex assembly—aiming to drive the large-scale industrial deployment of embodied AI.

"I believe a commercial rollout of humanoid robots in factories is achievable within two years," Zhang said.

At the ongoing 2025 World AI Conference in Shanghai, Galbot's robots showed live demonstrations, proving their operational maturity beyond the conceptual stage. Within a reconstructed supermarket environment designed to replicate real-world complexity, a robot executed unified cross-category handling, grasping diverse items from deformable snack bags to rigid bottles in cluttered shelves without pre-programmed paths.