PhDs from USTC take the lead in moving bricks! This listed company is actually the most secretive AI expert

Yang Jing from Aofei Temple
Quantum Bit Report | Public Account QbitAI

Loading and unloading is the oldest type of job that requires selling physical labor in exchange for pay.

The same thing happened in a factory in Hefei.

That day was a very ordinary rainy spring afternoon in Hefei, and the "workers" were busy moving all kinds of boxes.

There are cooking oil, milk, instant noodles, and all kinds of alcohol and tobacco...

Their daily task is to move the goods out of the containers and stack them into random stacks on pallets.

But what is different is that this group of "workers" is led by Dr. Zhang Jie from the University of Science and Technology of China, and their average age is only 28 years old.

Besides, you don't get paid just by moving one box every day.

What they are building is an intelligent loading system that relies solely on robots to run the entire process. It integrates multiple cutting-edge technologies such as robots, lidar, machine vision, digital twins, etc., and can independently complete the entire loading process including depalletizing, grouping, conveying, shaping, and stacking.

Currently, the system is nearing delivery and they are conducting the final round of debugging.

Carrying goods is a daily routine during the debugging process.

However, this cargo is a bit special - various cartons with mineral water weights , so that it can simulate cargo scenarios of various specifications and types at the same time.

What company does Dr. Zhang Jie work for?

Dr. Zhang Jie’s company - Taihe Intelligence .

Well, not often in the AI ​​news.

But it is an old player who has been deeply engaged in image recognition for 17 years.

Unlike other technology players who "find nails with hammers", Taihe has clear scenario requirements and aims at implementation from the beginning .

Taihe’s first scenario, which is also its entry point business, is called intelligent sorting .

Xu Dahong, chairman of Taihe Smart, recalled that this was the consideration at the time.

First, the technical threshold is high.

The color sorters used in the business are not only complex in design, but also involve a lot of optoelectronic knowledge. They also have very high requirements for software, hardware, and algorithms, and have technical barriers.

Customized designs are also required for different brands, such as vertical machines, crawler machines, tea machines, fruit and vegetable machines...

The second is the foreseeable implementation scenarios and market demand.

Breaking down the name, “Tai” means peace and stability, and “He” means food.

Since its establishment, Taihe has focused on the basic food of ordinary people - cereals and grains.

Rice, wheat, corn, peanuts, melon seeds... every grain of food we eat in our daily life may have been screened by Taihe.

Not every grain harvested from the fields can be eaten.

Unripe, moldy, contaminated...all reflected in the color.

If we were to select them one by one manually, it would take forever. However, Taihe used the principle of optics to teach machines to “select grains by color”, which greatly improved the quality of grains.

In 2012, sorting technology was further extended to tea leaves .

In 2018, even fruits were included.

Machine vision + AI can not only quickly identify bad fruits with injuries, but also grade the quality of fruits based on weight and size.

After using AI visual inspection, the machine can identify the type, color, beauty, ugliness, and number of wounds of the fruit at a glance, and then screen them.

The current screening range includes fruits such as apples, lemons, oranges, kiwis, and pears.

In addition to agricultural and sideline food sorting, Taihe is also involved in industrial fields such as plastics, non-ferrous metals, industrial salt, coal, and minerals.

Take AI ore sorting technology as an example.

In the past, coal was usually washed with water. Using centrifugal force, water can separate ores of different specific gravities, but this not only wastes water, but also causes environmental pollution.

But if combined with AI image recognition and X-ray imaging, the machine can automatically distinguish between coal and gangue, with an accuracy of 95%, and one ton of coal can save 70 kilograms of water.

Its public clients include many large energy groups such as Tongmei Group, Huaihe Energy, China Coal Group, and Shaanxi Coal Group.

At present, Taihe’s sorting business has covered more than 300 types of materials including grain, fruits and vegetables, food, industrial products, etc.

Xu Dahong revealed that this is not all, there will be new application scenarios, such as renewable resources.

This is what we often call garbage classification.

According to China News Service, citing a Japanese research institute, the total amount of garbage generated worldwide will reach 32 billion tons in 2050, 4.2 times that of 2000.

In addition to expanding its horizontal business around image recognition technology, Taihe has also further opened up the back-end factory manufacturing process.

And Smart packaging is the downstream of color sorting.

In 2019, Taihe established a subsidiary, Zhengyuan Packaging , which covers fields such as food, chemicals, seeds, and waterproof coatings.

The official website shows that the company’s current customers include Unilever (OMO), Procter & Gamble (P&G), Wilmar International, COFCO Group, and Sinopec.

Dr. Zhang Jie's intelligent loading system is a further step in the group's existing business scenario chain.

It is said that this was the most difficult project in the entire group at that time , with the largest investment, but it was the most determined one.

There were voices of opposition within the company, but Chairman Xu Dahong overcame all objections and supported research and development, believing that the power of technological innovation could bring a new installation model to the industry.

Four years later, this project was completed from scratch, and orders began to come in - simply because it met the rigid demand.

Among the first wave of customers are companies that are familiar to the general public, such as Langjiu, Yili, Golden Dragon Fish and other industry leaders.

Taihe has undergone three technical upgrades

From color sorting, intelligent sorting, to intelligent loading...

What is the driving force behind a scene-implementing company like Taihe, apart from its own pursuits, and what changes are taking place in the outside world?

Looking back on the development history, Taihe’s several business expansions were accompanied by product technology upgrades.

Xu Dahong revealed that Taihe has undergone three technology upgrades, and Dr. Zhang Jie's intelligent loading system was conceived during the second upgrade.

In 2004, Xu Dahong founded Taihe Intelligence, which engaged in grain sorting business with photodiode technology as the core.

(A photodiode, in simple terms, is a sensor device that converts light signals into electrical signals.)

A year later, the team developed an optoelectronic color sorter that uses light of a specific wavelength to separate objects based on differences in their reflectance and transmittance.

However, this equipment can only sort black and white materials and is limited to the belt channel trough, so the types that can be identified are limited.

In 2009, Taihe carried out its second technological upgrade.

The CCD color sorter they independently developed broke the technology monopoly of foreign companies and was the first domestic company to apply CCD image recognition algorithms in the field of grain sorting.

Compared with photoelectric color sorters, CCD color difference recognition rate is higher and it is more suitable for color sorting of various materials.

The momentum of this technological breakthrough was beyond Taihe's expectations.

Relying on this technology, Taihe has not only developed agricultural product sorting equipment such as grains, tea, fruits and vegetables, but also completed the listing.

In March 2017, Taihe Intelligent (then known as Taihe Optoelectronics) was listed on the main board of the Shanghai Stock Exchange.

This year, AlphaGo defeated Ke Jie in Wuzhen with a score of 3-0, and the AI ​​wave was revived in China again.

Taihe naturally did not miss this wave. It was also at this time that many cutting-edge projects, including Dr. Zhang Jie's intelligent loading project, began to be launched.

But compared to those AI companies that emphasize competition results and number of papers, Taihe is relatively low-key and also pursues implementation first.

This is related to the consistent philosophy of Chairman Xu Dahong.

He believes that technological trends represented by artificial intelligence will enrich industrial detection algorithms, provide new possibilities, broaden the boundaries of application, and, through the implementation of equipment, penetrate into the industry to achieve industrial upgrading.

We have also created the "TAI+ Intelligent Manufacturing Internet Platform" based on the integration of big data, AI, and Internet of Things technologies, an operation and maintenance platform serving customers that can provide analysis reports, virtual simulation processes, after-sales services, etc.

Dr. Zhang Jie's project stands out

Fortunately, Dr. Zhang Jie's intelligent loading project finally seized the opportunity of the times.

Now, intelligent loading is one of the four major businesses on Taihe’s official website.

Compared with the other three businesses, this new business is still a little "immature".

But internally it is already recognized as the future.

Because of its scene value.

The entire intelligent loading process is roughly divided into several steps.

The first step is parking positioning.

Depending on the different truck types, sizes, heights, levels of settlement (overall settlement when loading goods) and differences in the environment they are in, the loading robot needs to make real-time adjustments to ensure smooth cargo transportation.

The second step is to assign tasks. The entire system uses a digital twin system. Relying on digital modeling, workers can monitor the entire loading process online.

The third step is cargo detection. Image recognition is used to automatically detect the type, size, and placement of cargo on the pallet.

There are three main types of palletizing robots: box type, bag type and whole-pallet type. Different types correspond to different palletizing robots.

Everything is ready, and it’s time to load the truck.

The entire process includes depalletizing, grouping, conveying, shaping, and finally palletizing and stacking, which involves loading space planning algorithms, visual recognition and positioning algorithms, and autonomous decision-making technologies.

Through 3D visual recognition, the position of the carriage is accurately located to guide the stacking and loading of goods; relying on autonomous decision-making, the best way to unstack and stack goods is determined; when faced with different goods, algorithms need to be planned to find the optimal solution for loading in a limited space...

Dr. Zhang Jie introduced that this loading system has a certain versatility and can adapt to different trucks, goods, and complex environments. Therefore, it can be popularized in various industries and has broad application prospects.

In terms of cost-effectiveness, the system's bagging efficiency can reach 1,000 bags per hour, and its boxing efficiency can reach 1,800 boxes per hour . This can replace four or five workers at a time and maintain 24-hour operation. It is expected to help companies reduce their overall costs by 40% within five years.

Some time ago, the Human Resources and Social Security Bureau released a list of the top 100 occupations with the most labor shortages. Loaders and unloaders ranked 28th, which can be said to be a rigid demand for enterprises.

It not only reduces the factory's transportation costs, but also solves the company's labor shortage problem and avoids safety hazards during manual transportation.

Perhaps it is based on this that Zhang Jie has the confidence to make the bold statement that "I can make back my investment in two years".

Taihe has the confidence to make intelligent loading a key project to be promoted in the next 3-5 years.

There is such a plan map on Taihe’s official website.

Product sorting and processing area, intelligent packaging and palletizing area, intelligent storage area, and finally the intelligent loading area...

The entire back-end process is being fully opened up, and Xu Dahong's ideas are gradually being realized.

It is hard to imagine that a 17-year-old company has new products leading industry changes at every stage of development, has a continuous passion for innovation, and conducts forward-looking research for the next three, five or even ten years.

Taihe relies on nothing else but to go deep into the scenarios and then drive technology, deeply focus on an industry and thoroughly understand a vertical scenario.

This is also the aspect that is easily overlooked in the current implementation of AI - too many eyes are focused on AI algorithms and technological innovation unicorns, and on star companies holding hammers, but ignore the many "nails" in the scene. Enterprises, starting from niche fields, integrate AI into the research and development of smart equipment, and take the lead in realizing industrialization in a small but precise manner.

In this wave of AI revival, algorithms are the key, but the key is how to industrialize them. Taihe Intelligence is such a representative, a representative who quietly eats the dividends of AI.

There are many such companies, but they are very low-key. They are among the people, and they are quietly implementing their principles, reducing costs and increasing efficiency.

They also break the prejudice that AI is "incapable" of implementation from another perspective, and they will increasingly move to the center of the stage.

Do you know any other scenario AI companies like this?