Group Core Technology releases the SpatialVerse spatial intelligence platform to accelerate embodied intelligence and AIGC training | The Frontline

Written by Tian Zhe

Edited by Su Jianxun

From being able to only grasp objects to now learning to walk and run, the continuous evolution of the capabilities of humanoid robots is inseparable from the collection of data in real-world interactions. However, due to the high cost and slow speed of training in the real world, most robot companies adopt a form of real data + synthetic data to accelerate the training of humanoid robots.

However, although synthetic data can also simulate various situations in the real world, there are still differences in its quality compared to real data, which affects the training effect of robots.

On November 20, Qunhe Technology launched the Spatial Intelligence Platform SpatialVerse, which can provide synthetic data consistent with the material of real objects to accelerate the training of the robot's ability to grasp objects. Currently, Qunhe Technology has reached cooperation with many well-known domestic humanoid robot companies.

According to Chen Hang, the co-founder and CEO of Qunhe Technology, SpatialVerse has a massive deep learning data set for indoor scene cognition, and has the capabilities of simulating the real characteristics of the indoor environment, automatic segmentation and annotation, scene enhancement, and multi-platform docking, providing spatial intelligence training-related services for embodied intelligence, AIGC, and other fields.

It is reported that there are a wide variety of objects in SpatialVerse for customers to use, such as animal excrement, to facilitate the improvement of the robot's recognition and obstacle avoidance capabilities.

Qunhe Technology was founded in 2011 by Huang Xiaohuang, Chen Hang, and Zhu Hao, who are graduate students from the University of Illinois at Urbana-Champaign in the United States.

Chen Hang said that Qunhe is a technology company based on GPU clusters and AI technology, and has built a set of physically correct world simulators.

In the past few years, the company has focused on physical space simulation and applied it in scenarios such as real-time design rendering based on space, industrial production and manufacturing, and virtual physical world training. Up to now, Qunhe Technology has accumulated 320 million 3D models, with an average of 77.8 million monthly active visitors.

When it comes to the company's advantages, Huang Xiaohuang, the co-founder and chairman of Qunhe Technology, said that thousands of companies have used Qunhe Technology's tools for production and manufacturing. Qunhe generates tens of millions of sets of houses every year, and each component corresponds one-to-one with a real object. No company in the world can achieve this.

In order to expand this advantage, Qunhe Technology is building an ecological system. In the past year, its ecological system has added 450 new ecological applications, with a cumulative usage of over 20 million times, a year-on-year growth of 900%. In terms of industrial chain collaboration, Qunhe Technology will launch different versions of design and production docking solutions for enterprises of different scales.

In addition, Qunhe Technology has also released the Qizhen Engine for rendering, and the Matrix Engine with a multimodal CAD large model.

Source: Qunhe Technology

Chen Hang introduced that the Qizhen Engine is an end-cloud integrated ray tracing engine independently developed by Qunhe, which can help users map their imagination in the physical space and ensure its physical correctness.

Currently, the Qizhen Engine has been iterated to version 3.5, not only achieving real-time rendering in the cloud, but also achieving realistic-level rendering of the physical world. Combined with AI technology, this engine has overcome the problem of realistic rendering of organic matter, can render 99% of the materials in the physical world, and can simulate complex forms such as humans.

The Matrix Engine converts multimodal spatial information into physical space solutions through a multimodal CAD large model, and is ultimately used for production and construction.

It is reported that Qunhe will also customize enterprise-specific models based on the CAD large model for the special scenarios of enterprises and industries. Currently, Qunhe has achieved model customization cooperation with large home furnishing enterprises such as Gujia Home Furnishing.