#China #Shenzhen #ZEEKR #Humanoid Robots #UBTECH

Unleashing Swarm Intelligence: A New Dawn in Humanoid Robotics

In an impressive showcase of cutting-edge technology, UBTECH Robotics has launched the world's first collaborative training program for humanoid robots at ZEEKR's 5G Intelligent Factory. This groundbreaking initiative illustrates how multiple humanoid robots can seamlessly work together in industrial settings, effectively tackling a variety of tasks and operating in diverse scenarios. This leap forward marks a pivotal shift from individual agent autonomy to the sophisticated realm of swarm intelligence.

UBTECH's Walker S1 humanoid robots are at the forefront of this defining moment in robotics. Equipped with a networked cluster, they explore how advanced embodied intelligence can emerge through swarm behaviors and real-time interactions with their environment. This initiative is powered by BrainNet, UBTECH's innovative software framework facilitating collaboration among humanoid robots, alongside the Internet of Humanoids (IoH), which serves as a control center orchestrating the execution of swarm intelligence across the system.

This integration of BrainNet effectively links the collaborative inference nodes of cloud devices with competence nodes, creating a superbrain and a smart sub-brain within the swarm intelligence ecosystem. The superbrain operates on a large multimodal cognitive model that supports intelligent hybrid decision-making, guiding complex tasks throughout the production line, while the intelligent sub-brain, based on transformer models, integrates cross-domain fusion perception and collaborative control, allowing for parallel and distributed learning. This transformational capability equips humanoid robots to transcend beyond isolated task execution, enabling them to work synergistically at the production line level, ensuring efficient teamwork in complex industrial formats.

UBTECH has also developed the first multimodal model for humanoid robots, core to the superbrain, facilitating the continuous self-evolving nature of BrainNet. Currently being perfected using the DeepSeek-R1 technology, this model is designed to handle big data, granting humanoid robots human-like reasoning abilities. This means they can autonomously delegate, plan, and coordinate tasks, thereby optimizing the collaboration of multiple robots within intricate industrial workflows.

Progressive In-House Training: Multiple Robots, Multiple Scenarios, Collaborative Tasks

Now entering the realm of Practical Training 2.0, UBTECH's humanoid robots have evolved from single-agent autonomy to swarm intelligence. At ZEEKR's 5G Intelligent Factory, many Walker S1 robots are deployed across various complex production areas such as the final assembly workshop, PLC instrumentation section, quality inspection area, and vehicle assembly stations. They demonstrate seamless cooperation by undertaking collaborative sorting, handling, and precision assembly tasks, showcasing their effective teamwork in real-world industrial environments.

During the collaborative sorting phase, UBTECH's humanoid robots employ a vision-based perception technology and intelligent hybrid decision-making to enhance sorting activities. Utilizing purely visual, cross-domain perceptual systems, the robots can track dynamic targets in various environments, allowing for swarm collaboration through collective mapping and shared intelligence. Their intelligent hybrid decision-making system, based on an extensive multimodal model, effectively merges semantic VSLAM navigation with dexterous manipulation capabilities. This synergy enables dynamic task assignment between cloud-based and on-device brain nodes, fostering collaborative decision-making processes essential for swarm intelligence.

In collaborative handling scenarios, these humanoid robots face several challenges, such as uneven load distribution, complex trajectory planning, and dynamic environmental adaptability. To address these, UBTECH has devised a shared planning and control system that coordinates multiple robots in trajectory planning, load detection, and compliant control. This guarantees that the robots can dynamically adjust their posture and force during handling, significantly enhancing stability and efficiency when transporting large, heavy items.

In precision assembly tasks, Walker S1 robots display exceptional dexterity, particularly when dealing with deformable materials. While working with small and pliable film materials, they utilize high-precision sensors and adaptive control technologies to dynamically adjust their grip strength and posture. This ensures that film objects remain intact and properly aligned throughout the assembly process, emphasizing the reliability and versatility of UBTECH’s humanoid robots equipped with tactile sensor-based dexterous hands suitable for intricate industrial applications.

Furthermore, the Walker S1 incorporates vision-based global positioning with a force-based secondary positioning via reinforcement learning, enabling them to execute operational quality inspection tasks with unmatched precision and adaptability.

UBTECH is collaborating with several leading companies in the automotive industry, including Dongfeng Liuzhou Motor, Geely Auto, FAW-Volkswagen, Audi FAW, BYD, BAIC New Energy, Foxconn, and SF Express. The Walker S humanoid robots are now more widely deployed across automotive manufacturing facilities than any other humanoid robot, following the success of the first phase of training for autonomous single agents. With the launch of Practical Training 2.0, UBTECH is expediting collaborative training for multiple humanoid robots and advancing AI model development to drive product and technology innovation. Besides scaling up with Geely Auto, UBTECH is also extending its reach into various partner plants, reinforcing the scalable use of humanoid robots across industrial applications.

For further insights, check out this video overview.