Former Meituan Waimai Head Enters Restaurant Embodied AI, AtomBite.AI Raises Millions in Seed Funding

Embodied AI startup AtomBite.AI (元节智能) has completed a multi-million yuan seed funding round, led by InnoAngel Fund with participation from Tsinghua Alumni Seed Fund and individual investors. The company, registered on March 31, 2026, has already secured product deployment intentions from multiple leading domestic and international companies, even before mass production.

Founder Dr. Wang Dong, former technical head of Meituan Waimai (the food delivery arm of Meituan), led a team of over 1,000 in R&D and product development, building the algorithm, data, and system architecture that supports tens of millions of daily orders. His academic background includes a PhD under Professor Zhang Bo at Tsinghua University, and he developed the world's first commercial video face recognition and tracking system in 2011. Co-founders Li Tao, another Meituan veteran who oversaw the algorithm and data systems for Meituan Waimai, and Li Haozhe, a serial entrepreneur with global commercial experience, round out the team. Other core members come from Tsinghua, USTC, Meituan, Horizon Robotics, and more.

Rather than chasing the crowded humanoid robot space, AtomBite.AI is focusing on the "less sexy" yet highly practical domain of restaurant back kitchens. After leaving Meituan, Wang Dong spent months visiting restaurants and platforms in North America and Singapore, concluding that the kitchen is the most deterministic commercial direction for embodied AI. The global food delivery order volume continues to grow, but the link between meal preparation and delivery handoff remains a technological weak spot with high error rates and spillage, costing all parties involved. Additionally, the restaurant industry faces structural labor shortages.

The company's key innovation is its "embodied world model," which moves beyond the reactive VLA (Vision-Language-Action) paradigm towards a World Action Model (WAM). This approach trains the model to predict the outcome of actions before executing them, enabling complex planning in chaotic environments like kitchens. Three-layered architecture: top layer is the embodied world model for scene understanding and action prediction; middle layer is dual engines for task orchestration and scheduling optimization; bottom layer integrates self-developed core components with general hardware. The commercial path starts with takeout packaging and docking to reduce order errors, then extends to controlling existing kitchen equipment like cooking robots and dishwashers, ultimately building a digital twin kitchen operating system.

Seed funding is just the beginning for this company less than two months old. With a strong team, clear technology route, and a business model grounded in real-world data collection from thousands of restaurant kitchens, AtomBite.AI aims to lead the era of embodied application starting from the back kitchen.