Exclusive: Boundless Dynamics Raises Over $70 Million to Build a 'Universal ...

Boundless Dynamics, a Chinese startup developing embodied artificial intelligence (AI) robots, has raised about $42 million in an angel financing round co-led by Sequoia China and Linear Capital, according to people familiar with the deal.

Other backers include Hillhouse Venture Capital, Horizon Robotics, Hua Capital, Eastern bell Capital, Baidu Ventures, and Tongge Ventures.

The company is close to completing an “Angel+” round, bringing its total fundraising to more than $70 million, the people said.

Founded in 2025 by Zhang Yufeng, a former vice president of Horizon Robotics, Boundless Dynamics is focused on developing a “universal operating brain” for robots — a foundational model that integrates perception, cognition, and control. The company aims to make embodied intelligence a scalable infrastructure that can be widely deployed across industries.

“The new funding will significantly accelerate our dual-track strategy of foundational model R&D and expert model deployment,” Zhang said. “This will allow us to create immediate value for customers while advancing the long-term evolution of a universal operating brain.”

Boundless Dynamics’ robotic platform has already achieved technical breakthroughs in industrial manufacturing and commercial services. The company is now working with domestic and international partners to deploy its systems in real-world settings, with formal factory rollout expected as early as this month.

Embodied AI — the integration of artificial intelligence with physical robotics — is entering a crucial stage, shifting from research validation to commercialization. Industry experts widely believe that achieving fully general, cross-domain intelligence will take at least another decade.

Currently, the task success rate of large-scale multimodal embodied models in industrial scenarios remains below stable commercial standards — a major hurdle for practical deployment.

Instead of pursuing an all-in-one “super model” immediately, Boundless Dynamics is taking what Zhang calls a “pragmatic path” — achieving “industry-level generalization” and large-scale deployment before expanding to cross-industry capabilities.

That approach requires multidimensional innovation across algorithms, data, and engineering.

Building a robot that “understands” the physical world

At the core of Boundless Dynamics’ R&D is the development of a foundational embodied model capable of perceiving and reasoning about the real world. The company’s scientists aim to build systems that can “understand” physical laws, track changes in three-dimensional environments, and perform real-time reasoning and decision-making in complex, dynamic scenarios.

Ultimately, Zhang said, the goal is for the model to “learn how to learn” — an AI that can continuously evolve through real-world interaction.

The company’s data strategy combines simulated data with “Three Real” datasets — real people, real machines, real environments. By collecting operational data from live factory production lines and commercial environments, Boundless Dynamics bridges the gap between simulation and reality.

This allows its models to maintain stable performance in complex tasks such as object grasping, component assembly, and flexible material handling — even under conditions like environmental shifts, occlusion, or spatial constraints.

Boundless Dynamics operates on a dual-track system that integrates foundational model research with the deployment of commercial-level “expert models.”

These expert models — optimized for specific industrial and service scenarios — generate high-quality feedback and interaction data that flow back to improve the foundational model. In turn, upgrades in the foundational model enhance the performance of expert models, creating a self-reinforcing learning loop.

This dynamic cycle, Zhang explained, allows the company to build a complete chain from model R&D to application deployment and capability evolution — “learning through application, and evolving through learning.”

To validate and iterate its models in real-world conditions, Boundless Dynamics has introduced a “Shadow Mode”mechanism borrowed from autonomous driving technology. The system lets the foundational model operate silently alongside the expert model during live tasks, gathering error cases and boundary data without disrupting production.

That feedback is then fed back into the model training system, accelerating the model’s ability to adapt to real-world environments.

On the engineering front, Boundless Dynamics is building a deeply integrated software-hardware platform that combines embodied intelligence models, computing infrastructure, actuators, and real-time operating systems.

Leveraging its founder’s experience in automotive-grade engineering from Horizon Robotics, the company aims to ensure stability and reliability across diverse real-world scenarios — turning lab-tested algorithms into industrial-grade solutions.

The technical architecture can be summarized as a closed-loop ecosystem:

1. Foundational models construct deep cognition of the physical world.

2. Real-world “Three Real” data enhance model success rates and generalization.

3. Massive computing power and the Shadow Mode mechanism enable continuous evolution.

4. Integrated engineering ensures deployment stability.

Through this structure, Boundless Dynamics hopes to efficiently bridge the gap between AI research and scalable, real-world embodied intelligence systems.

A veteran founder with deep AI roots

Zhang Yufeng brings a cross-disciplinary background that spans both technology and business execution. Before founding Boundless Dynamics, he worked in R&D management roles at Sony and ARM, and later joined Horizon Robotics in 2017, where he led a 1,000-person team in intelligent driving software development, engineering, and mass deployment.

Under his leadership, Horizon Robotics achieved the top market share for advanced driver-assistance systems (ADAS)among Chinese passenger car brands and secured strategic partnerships with BYD, Changan Automobile, Volkswagen Group, and Continental Group.

Boundless Dynamics’ co-founder and CTO, Xu Wenda, holds a Ph.D. in robotics from Carnegie Mellon Universityand has extensive experience in autonomous driving and robotics entrepreneurship.

During his doctoral studies, Xu co-founded Ottomatika, an autonomous driving startup in the U.S., and led the launch of the world’s first commercial self-driving ride-hailing operation. He later served as Vice President and Chief Algorithm Architect at Carnegie Power Autonomous Driving.

The Boundless Dynamics team also includes scientists who have made milestone contributions in multimodal large models, reinforcement learning, and world models, as well as an engineering team with experience in mass production of millions of software and hardware systems.

Investors say Boundless Dynamics’ combination of foundational model R&D and industrial execution gives it an advantage in the fast-growing embodied AI field.

Zhang Han, Partner at Sequoia China, said the firm has long been tracking embodied intelligence and believes Boundless Dynamics has “the rare ability to connect technology, product, and scale” while the field is still evolving.

“The team led by Yufeng has both deep technical innovation and strong engineering execution,” Zhang said. “We expect them to become a major force driving progress in this industry.”

Wang Huai, Founder and CEO of Linear Capital, praised Zhang’s full-stack experience, noting that he is “one of the few who has personally gone through every stage from technology to product to market to scale.”

“What sets Boundless Dynamics apart,” Wang said, “is its focus on anchoring cutting-edge embodied intelligence in real-world scenarios, turning the universal dream of ‘reaching for the stars and the sea’ into tangible productivity.”

Yu Kai, Founder and CEO of Horizon Robotics, described Zhang as “a longtime partner and core leader” who helped deliver over ten million intelligent driving systems during his tenure at Horizon.

“Now he’s applying that same systematic innovation to embodied intelligence,” Yu said. “We’ll continue to support Boundless Dynamics as it accelerates the arrival of the era of general embodied intelligence.”