UBTECH Completes First Swarm Intelligence Training for Humanoid Robots at ZEEKR Factory

UBTECH Robotics has completed the world's first collaborative practical training program for humanoid robots at ZEEKR's 5G Intelligent Factory, where multiple UBTECH Walker S1 humanoid robots worked together on the assembly line.

According to UBTECH, this demonstration proved that the industry is inching closer to a technology called swarm intelligence. Unlike traditional single-agent autonomous robots, UBTECH's Walker S1 humanoid robots operate as a coordinated swarm, controlled by the centralized UBTECH BrainNet AI network. By transitioning from single-agent autonomy to multi-robot systems, the swarm can better handle complex manufacturing tasks, such as material sorting and precision assembly.

The robots operated using UBTECH's BrainNet framework, which integrates cloud-based and on-device intelligence for collaborative decision-making. The system's main controller (called a "super brain"), powered by a large reasoning multimodal AI model, manages complex tasks, while a sub-controller (called a "sub-brain") enables real-time multi-robot coordination. This hierarchical system allows the robots to adapt quickly to changing workloads and conditions in industrial settings.

Furthermore, UBTECH's large reasoning multimodal AI model, based on DeepSeek-R1, also processes data at scale to bestow the robots with "human-like common-sense reasoning." The model, which is currently under development, incorporates Retrieval-Augmented Generation (RAG) technology, allowing it to adapt to specialized job functions and improve decision-making accuracy and scalability.

During the training process, the Walker S1 humanoid robots were deployed across several ZEEKR production segments, including final assembly, quality inspection, and vehicle assembly. They used a multi-sensor system to perform a variety of tasks. For example, during the sorting process, the robots relied on vision-based perception and hybrid decision-making to track and categorize items. For material handling, they managed heavy and uneven loads by adjusting their posture and force distribution. Finally, during assembly, the robots used high-precision sensing and reinforcement learning to perform and inspect finished products, even for small and deformable materials.

This wasn't UBTECH's first foray into swarm intelligence. The company had previously piloted its technology in several partner EV factories to build out a sustainable business model. Then, as industrial automation grew more complex, UBTECH shifted its focus from single-agent control to swarm intelligence to meet the industry's demands for operating humanoid robots at scale.

Beyond ZEEKR, UBTECH is collaborating with automotive manufacturers such as Geely Auto, FAW-Volkswagen, Audi FAW, BYD, Foxconn, and SF Express. The company recently rolled out its Practical Training 2.0 program, which focuses on advancing its Swarm Intelligence technology in industrial automation.

Although this round of swarm intelligence training was conducted with the UBTECH Walker S1, the company is already preparing newer versions, including the Walker S2 and Walker S3, for release later this year. Regardless of labor costs, UBTECH believes that swarm intelligence will fundamentally transform the way manufacturing operates.