“To enter the real world, AI must first learn three essential skills: sensing physical reality, understanding spatial environments, and creating virtual worlds”

This was a shared consensus reached by technology scholars and industry experts at the 2026 2nd ASIASIM International Summer School. As large-scale models continue to advance rapidly in natural language processing and 2D image generation, a key challenge at the intersection of simulation science and artificial intelligence has emerged: how to bridge the “virtual–physical singularity” and enable embodied intelligence to achieve accurate perception and autonomous decision-making in the three-dimensional physical world.

During the closing ceremony of the summer school, three innovation representatives — Zhixiang Wang, Head of University–Enterprise Cooperation at the Education Division of Keenon Technology; Wenyun Feng, Vice President of Human Resources at LingoAce Technology; and Zhengye Wang, Education Business Director of the Cultural Tourism Division at Rokid Technology — delivered highly informative thematic presentations.

From different perspectives, they showcased the diverse future directions of simulation technology. Through their respective technological breakthroughs, the three companies revealed three critical dimensions of next-generation simulation technology and demonstrated its evolution as a fundamental infrastructure for AI.

Core Framework: A Technology Roadmap for Next-Generation Simulation

To clearly illustrate the technological logic, these innovations can be summarized into three key technological layers that AI must overcome on its journey into the physical world:

01 LingoAce Technology (Style3D)
Physical Perception — Flexible Body Dynamics Modeling: The Foundation for Embodied Intelligence

Apple Drop PhysicRender

In the field of digital twins, rigid body simulation has become highly mature. However, deformable body simulation (such as cloth, hair, and soft tissues) remains a world-class challenge in computer graphics and physics due to the complexity of real-time collision detection and deformation calculations involving millions of vertices. By overcoming this critical challenge, LingoAce Technology has established a strong technological advantage in the field of physical simulation.

Industrial-Grade Flexible Physics Engine

LingoAce’s independently developed core engine enables highly precise microscopic simulations of flexible materials, such as fabric fibers, by calculating multidimensional mechanical properties including stretching, bending, and shearing. At the macro level, it achieves high-frame-rate real-time rendering, reducing long-term dependence on overseas commercial simulation software in this field.

Cross-Domain Pipeline Expansion

With its highly realistic and efficient physical simulation capabilities, LingoAce’s technology has expanded beyond the apparel manufacturing industry. By providing physics simulation plugins for mainstream digital content platforms such as Unreal Engine and Maya, the company has become deeply involved in CG production, digital human development, and visual effects pipelines of major technology companies.

The “Tactile” Foundation for Embodied AI

As AI robots move from laboratories into real-world environments, a major challenge lies in their limited ability to perceive and interact with flexible objects, such as folding clothes, grasping soft materials, and navigating around deformable objects. By integrating flexible physics engines into simulation-based training environments, robots can learn to perceive and respond to different material properties in virtual worlds before deployment, completing a critical missing component of physical intelligence.

02 Rokid Technology
Spatial Understanding — 6DoF Tracking and Native Spatial Interaction

Large language models (LLMs) are advancing rapidly in intelligence, but if AI can only interact through typing on a smartphone, it will remain merely a “passive responder”. Rokid is enabling AI to truly “gain eyes and ears” through its industry-leading AR hardware and spatial operating system, allowing AI to achieve active contextual awareness of the physical world.

Technical Advantage

By integrating advanced SLAM (Simultaneous Localization and Mapping) point-cloud technology, the system achieves centimeter-level accuracy in 6DoF (Six Degrees of Freedom) spatial tracking, enabling virtual information to be precisely anchored and seamlessly integrated with real-world environments.

Deep AI Integration

Through deep integration with leading large language models such as DeepSeek and Tongyi Qianwen, Rokid’s AR devices have evolved from simple display terminals into intelligent spatial agents. Equipped with cameras and microphones, the devices can “see” historical architecture, “hear” multilingual inquiries, and overlay real-time navigation and translation subtitles onto the user’s field of view — creating experiences beyond what software alone can achieve.

Industry Applications

Rokid’s technology has been deployed across more than 300 major museums and 5A-level scenic attractions in China. Its business model extends beyond hardware, using AR terminals to activate historical and cultural data and create city-level digital cultural tourism ecosystems integrating intelligent guided tours, immersive digital storytelling experiences, and multi-terminal offline services.

03 ManyCore Technology

Environment Creation — Spatial Language Models and Synthetic Data Generation

Distance and Direction Awareness

When ManyCore Technology is mentioned, the public’s first impression is often “Kujiale”, a popular digital home design platform. Through this journey, ManyCore has accumulated massive amounts of physically accurate and interactive 3D spatial data. Led by three founders with strong backgrounds in computer science, the company has further advanced into the frontier of Spatial Intelligence.

Spatial LM and 3D Gaussian Splatting

ManyCore’s open-source Spatial LM (Spatial Language Model) has ranked among the top three on the Hugging Face leaderboard, following DeepSeek. By integrating 3D Gaussian Splatting, the company has transformed the traditional time-consuming manual mesh modeling workflow. Taking the reconstruction of ancient architecture featured in Black Myth: Wukong as an example, traditional 3D modeling may require nearly two months, while ManyCore’s spatial intelligence platform Aholo can complete scan-based reconstruction within two days, producing fully explorable and interactive 3D environments.

Synthetic Data and Robot Training Environments

Robotics models often lack an intuitive understanding of 3D spatial structures, physical rules, and geometric relationships. Through high-precision spatial reconstruction, ManyCore can generate large-scale 3D virtual environments equipped with physicalcollision boundaries, spatial semantic labels, and surface friction parameters, including digital homes and “lights-out” factories. Within these highly realistic simulation environments, robots can conduct millions of reinforcement learning (RL) trials, reducing the costly trial-and-error process of real-world deployment and helping overcome challenges such as robots failing when entering unfamiliar environments.

Industry–Academia–Research Ecosystem Development

To establish a systematic knowledge framework for spatial intelligence, ManyCore has collaborated with universities such as ZhejiangUniversity to launch a 64-hour SpatialIntelligence curriculum. Covering conceptual frameworks, technical systems, and practical platforms, the program aims to cultivate the next generation of interdisciplinary talent in simulation and AI.

From “Data-Driven” to “Data + Physics Integration”

The industry–academia–research discussions at the ASIASIM International Summer School revealed a clear technological trend: the evolution of intelligent simulation is undergoing a fundamental paradigm shift, accelerating from purely “data-driven” approaches toward the integration of “data + physics”.

For AI to truly enter the physical world, it requires more than breakthroughs in individual algorithms. It is a systematic engineering effort involving the simulation of physical laws, native spatial perception, and the generation of high-fidelity 3D environments.ASIASIM will continue to serve as an international platform connecting industry, academia, and research communities, bringing together global academic expertise and cutting-edge industrial innovation to advance breakthroughs and practical applications in next-generation intelligent simulation technologies.