HomeArticle

Raised 1 billion yuan within 10 months, covering 70% of leading embodied intelligence enterprises: Xingyuan Intelligence uses heterogeneous multi-robot collaborative rope skipping to demonstrate the "cross-ontology" capability of the embodied brain.

氪报2026-09-22 10:18
Xingyuan Intelligence showcases a number of technological achievements in embodied intelligence and promotes their practical application.

The embodied brain track sees growing popularity, Xingyuan Intelligence lands multiple technologies in practice

 

With the theme of "Embodied Brain · Interact with the World", Xingyuan Intelligence focuses on real tasks including industrial operation, dynamic interaction and heterogeneous collaboration, and centrally displays the cutting-edge technologies and application progress of embodied brain, world model and multi-robot collaboration.
 

From single-agent intelligence to multi-agent collaboration, multi-form robots perform "long rope skipping" on site

At this year's World Robot Conference, two humanoid robots collaboratively shake the rope, and the quadruped robotic dog completes continuous jumps following the rhythm of the long rope, which intuitively presents the collaboration capability between different robot subjects. This demonstration is jointly developed by Xingyuan Intelligence and the LAMDA team of Nanjing University. Based on the original RoboBrain Pro embodied brain, the team developed a heterogeneous embodied collaborative learning framework, which realized the world's first real-machine verification of heterogeneous multi-robot collaborative long rope skipping. Through hierarchical collaborative learning and closed-loop feedback, stable long rope control and cross-subject action synchronization are achieved; relying on multi-agent reinforcement learning, different robot subjects can independently learn collaboration strategies in continuous interaction.

 

From a single robot completing preset skills to different forms of robots jointly completing complex tasks, heterogeneous multi-robot collaboration further requires different entities to form real-time cooperation at the perception, decision-making and action levels, promoting the evolution of embodied intelligence from single-agent skill learning to multi-agent collaborative decision-making.
 

The embodied brain enters diversified scenarios, and real tasks verify intelligent capabilities

In addition to heterogeneous multi-robot collaboration, Xingyuan Intelligence also presents the application progress of embodied brain and world model in different entities and different tasks through on-site demonstrations such as industrial operation and high-speed dynamic interaction.

In the industrial scenario, the high-position operation robot "Optimus" jointly built by Xingyuan Intelligence and Zhongli Co., Ltd. is exhibited to the public for the first time. Equipped with the RoboBrain Pro embodied brain, it can achieve high-position operations at a maximum height of 10 meters. The embodied loading and unloading solution jointly built by the two parties has realized 90 seconds for single-vehicle loading and unloading, and 1 minute for two-vehicle collaborative loading and unloading, which keeps the same speed as manual operation.

In the dynamic interaction scenario, the robot equipped with the ω-EVA embodied interactive world model plays table tennis with on-site audiences, continuously perceives, predicts and adjusts actions in a high-speed changing environment, which intuitively presents the real-time interaction capability after the world model is incorporated into the closed loop of robot action decision-making.

The technical matrix is continuously improved, with open source and hardware advancing in parallel

Xingyuan Intelligence has intensively released three important achievements: heterogeneous multi-robot collaboration demonstration, open source world model JEPA-WAM, and N5C liquid-cooled edge-side computing power platform, demonstrating the technical progress of embodied intelligence from demonstration Demo to industrial implementation. JEPA-WAM adopts the Shared Predictor architecture, with only 0.5B parameters for the base model, no massive robot training data is required, which reduces the computing power and data threshold for world model deployment, and forms technical complementarity with the previous ω-EVA embodied interactive world model. The new N5C liquid-cooled edge-side computing power platform is equipped with the THOR T5000 computing core. Its liquid cooling heat dissipation performance can reach 3-15 times that of traditional air cooling, supports stable operation in a high-temperature environment of 70℃, and has IP56 protection level.
 

Reshape the new paradigm of industrial collaboration through systematic collaboration

From industrial production, high-speed dynamic interaction to heterogeneous robot collaboration, what Xingyuan Intelligence demonstrates is not just a single robot skill, but the capability of embodied brain and world model to span different entities and adapt to different real tasks. Liu Dong, founder and CEO of Xingyuan Intelligence, said that large models and high computing power are both means to boost productivity. The industrialization of embodied intelligence needs to take into account the three core indicators of effect, efficiency and cost, and the edge-side brain will be the inevitable path for the large-scale implementation of advanced embodied intelligence.