Why does a giant in the clothing technology field enter the field of embodied intelligence?

The golden age of "Physical AI" may be approaching.

At the end of April, GroupCore Technology went public on the Hong Kong Stock Exchange, and its stock price soared by nearly 400% in two days. This leading company in the home furnishing technology track has accumulated a vast amount of 3D model data with real physical parameters over 15 years. It successfully became the first stock among the "Six Rising Stars in Hangzhou", officially kicking off the era of "Physical AI".

Interestingly, also in Hangzhou, another industry giant is doing something similar.

However, different from GroupCore Technology, which focuses on "3D rigid body simulation", it is targeting the "3D flexible simulation" track, which has extremely high technical difficulty and extremely scarce data.

In February this year, Lingdi Technology released its self - developed physical simulation and synthetic data system, SynReal. As a giant in the clothing technology track, Lingdi Technology has accumulated a vast amount of "3D flexible data" of fabrics in 10 years. Its SynReal system can reduce the training cost of embodied intelligent simulation and improve the data throughput at the same time. For example, the Spring Festival Gala clothes - folding robot of Galaxy General has partial technical support from Lingdi Technology.

Galaxy General's Spring Festival Gala clothes - folding robot (Image source/Enterprise)

Why is "flexible simulation" so difficult? Why is it called the "Holy Grail of Embodied Intelligence"? Why does a giant in the clothing technology track choose to cross into the robot track, and what does it rely on to succeed?

The story starts with the "problem of robots folding clothes".

01 Robots can't even fold clothes

When you think that an embodied intelligent robot is "smart", try asking it to fold a piece of clothing.

This cutting - edge technological creation, which can carry heavy loads, run long distances, and even outperform humans in half - marathons, may be at a loss when facing an ordinary T - shirt, unable to flatten it, grasp it firmly, or fold it neatly.

Of course, there are also many players who can "fold well", and there are generally two implementation paths:

One is to "fix" the action program. The clothes are placed in a specified position, flattened in advance, and then the machine performs fixed actions;

The other type of robot is "smarter". It can flexibly grasp different clothes and continue the task even in the face of human interference. However, the cost is that it needs to collect real - human operation data for more than half a year as the "training materials" for the robot. Once the scene, lighting, desktop, or clothing material is changed, the "materials" cannot be reused.

Smart as you are, you must be thinking, can we build a "virtual world" that conforms to the physical rules of the real world in the computer and train the robot with code?

Indeed, many giants are doing this, such as NVIDIA Issac, Microsoft AirSim, and MuJoCo acquired by DeepMind. Different from AI software such as large models, since this path is deeply bound to the physical world, it is often called "Physical AI".

However, as mentioned above, in the current Physical AI, "rigid data" dominates. When you open the database, it is full of cubes, robotic arms, and regular objects. They are easy to define and calculate.

But we all know that in the real world, from the clothes we wear to the fruits in the supermarket; from the plastic bags we knead casually to our own skin... at least one - third of the objects are "flexible".

The movement trajectory of a rigid object is relatively certain. However, for a piece of soft cloth, each fiber is fluttering freely.

When the robot grabs it and applies a slight force, it will trigger a chain reaction. How do the wrinkles form? How does the cloth hang? The change of each point is related to the whole. An ordinary piece of cloth is discretized into tens of thousands of vertices in 3D simulation. Each vertex has multiple degrees of freedom. This will bring an astronomical amount of calculation, and the difficulty soars.

Lingdi Technology's flexible 3D simulation technology (Image source/Enterprise)

When two flexible objects come into contact, such as when folding clothes, it is the folding and contact of the cloth with itself. The complexity of the physical interaction is simply "hell - level".

This is why "flexible 3D simulation" is called the "Holy Grail of Embodied Intelligence". It not only has extremely little data but also has high calculation difficulty. However, it is a technical problem that must be overcome for the implementation of embodied intelligence.

After all, imagine what terrible consequences it would bring if an elderly - care robot treats humans as "rigid bodies" that do not deform.

02 A "hidden master" emerges from the clothing industry

The problem of "flexible 3D simulation" has troubled the industry for many years. When people focused the spotlight on robot companies and AI laboratories, no one expected that the real game - changer would come from a seemingly "unrelated" field.

3D clothing technology.

How does a virtual piece of cloth hang? How does it flutter? How does it entangle, bend, stretch, and fold with another piece of cloth? These problems, which are regarded as "hell - level" in the field of physical simulation, are exactly the daily work of Lingdi Technology.

This clothing technology giant, founded in 2015, has accumulated a vast amount of high - quality and extremely scarce "3D flexible simulation data" in ten years. As an invisible champion in the clothing technology field, it has not only been on the "List of Pre - Unicorn Enterprises in Hangzhou" for a long time but also ranked among the "Eighteen Arhats of Embodied Intelligence in Hangzhou" with its flexible 3D simulation technology.

Of course, just having data accumulation is not enough.

In the past ten years, Lingdi Technology's research team has made amazing long - term investments in the basic research of physical simulation of flexible objects. It has published many research results at top global graphics conferences such as SIGGRAPH, covering core technical issues such as physical simulation of deformable bodies, complex contact processing, and high - performance numerical calculation.

Lingdi Technology's paper results in 2025 (Image source/Enterprise)

So, when the embodied intelligence industry is struggling for half a year to collect data for "folding a piece of clothing", Lingdi Technology has already stocked up ten - year's worth of ammunition in its "arsenal". When people are suffering from the "hell - level" difficulty of flexible 3D simulation calculation, Lingdi Technology's basic research is "looking for solutions" to this problem.

Now, this "hidden master" finally steps into the spotlight.

The SynReal physical simulation and synthetic data system launched by Lingdi Technology is the "masterpiece" that condenses its ten - year efforts.

It will try to hold up that "Holy Grail" and let robots truly touch the "softness of the world".

03 3D simulation determines the upper limit of embodied intelligence

SynReal is mainly composed of three parts:

1. SynReal Sim - a high - fidelity simulation engine;

2. SynReal Arena - an embodied intelligence training platform;

3. SynReal Core - a training model based on large - scale synthetic interaction data.

It may sound a bit complicated. To put it simply, SynReal Sim is responsible for creating a "virtual world" that conforms to physical rules, SynReal Arena provides a "virtual training ground" for robots, and SynReal Core is responsible for enabling robots to learn and develop abilities.

Combined together, they form a SynReal "robot training school". Robots can gradually learn how to interact with the physical world through millions of "virtual trainings" per minute in it. After "graduating" from the school, they can "work" in the real world.

Compared with the "virtual school" of physical simulation and synthetic data systems like SynReal, another training idea for embodied intelligence is to train robots through manual data collection, the so - called "People's Education Edition" robots.

In comparison, this training method is simpler in terms of technical implementation. Since the "People's Education Edition" data comes from the real physical world, it can contain certain details and noise, which can help robots adapt to the real environment.

However, the disadvantages of the "People's Education Edition" are also extremely obvious - it is expensive, very time - consuming and labor - intensive, and difficult to generalize.

You know, the collection time of the "People's Education Edition" often takes "months" or "years" as the unit. Collecting high - quality demonstration data for a single task often takes a professional team several months, and the cost ranges from hundreds of thousands to millions of yuan.

The simulation platform can generate millions of data trajectories with rich variations within a few hours, and this cost control and data scale are almost impossible tasks for the "People's Education Edition".

Compared with other physical AI simulation platforms, "real, fast, and stable" are the three unique advantages of SynReal.

First is "real". With the support of a vast amount of rigid and flexible 3D simulation data, SynReal can not only provide a more real, complex, and real - world - like training environment for robots but also achieve more accurate and correct static and dynamic calculations. Its error is nearly 20% less than that of the industry benchmark, Issac Sim.

High - fidelity physical simulation effect (Image source/Enterprise)

Second is "fast", which is also one of the most important indicators of the "robot virtual school". The simulation speed determines whether data generation is feasible on a large scale. Lingdi's technical team has reconstructed the overall process of "flexible 3D simulation" around the GPU parallel computing architecture, which can greatly improve the simulation throughput. The simulation speed is 5 - 10 times faster than that of Issac Sim, significantly improving the robot training efficiency.

Finally is "stable". As mentioned above, 3D simulation of a single flexible object is already very difficult, and the physical interaction difficulty when two or more flexible objects come into contact is simply "explosive". In 3D games, flexible objects such as hair and clothes are also most likely to "pass through models". To solve this problem, Lingdi's technical team introduced a path calculation method based on IPC (Incremental Potential Contact, the latest innovative idea at SIGGRAPH 2020) in the simulation, enabling SynReal to maintain simulation stability even in the case of frequent changes in multi - point, multi - layer, and self - contact, allowing robots to obtain more stable and reliable learning results in complex scenarios.

Only a more realistic, efficient, and stable "virtual school" can train more intelligent, dexterous, and adaptable embodied intelligent robots.

At present, when labor costs are becoming increasingly expensive and the demand for robot training is soaring, using a vast amount of low - cost and high - quality simulation data to train robots is undoubtedly the general trend.

Only through the tempering of the "virtual school" can robots truly reduce costs, change from expensive "laboratory treasures" and "stage performers" to "social labor forces" that can be deployed on a large scale, and enter families and all industries to do laundry, cook, take care of the elderly, and provide companionship.

Historical turning points are often driven by "cross - border players".

Just as NVIDIA's "original business" is game graphics cards and Xiaomi's original business is "enthusiast mobile phones", many technological breakthroughs and industry leaders are actually "cross - border players".

"Cross - border players" carry completely different industrial genes. They use their deep accumulation in new fields to solve stubborn pain points in old fields in a "leapfrog" way. They have no old - fashioned thinking patterns but only new tools and perspectives for problem - solving.

Flexible 3D simulation is the key technological watershed in the current embodied intelligence industry, and Lingdi, which owns a whole treasure trove of flexible 3D models, is exactly the "cross - border player" that the industry needs most at present.

Lingdi Technology's 10 - year accumulation of flexible 3D simulation technology in the clothing technology field, combined with its engineering capabilities deeply optimized for parallel computing, constructs a SynReal "virtual training ground" that is both extremely realistic and highly efficient and stable, directly hitting the old pain points of "expensive, slow, and difficult to generalize" in the robot training industry.

It is undeniable that today's robots are still too far away from people's lives.

In the past ten years, it has been a decade of rapid development for software AI. We have witnessed the emergence of neural networks, deep learning, and the globally popular large language models.

Just as the well - known online joke goes - "I hope AI can help me do the laundry and wash the dishes so that I can have time for art and writing. Instead of AI replacing me in art and writing while I do the laundry and wash the dishes."

Perhaps in the next ten years, whoever can master "Physical AI" will be able to define the next intelligent era.