Mobile Robots Come To A Fork in The Road

The mobile robotics (AGV/AMR) industry has been in the news lately. Purdue crossed the border from service robots into the industrial market, adding a new player to the already crowded track. Haikang Robotics announced that the 100,000th AMR product in all categories came off the line, the industry's head player after years of development, production capacity is taking off, the scale of the development trend has appeared.

At the same time, since last year's big model boom, also ushered in a new development of mobile robotics industry. One entrepreneur excitedly said that "it improves the level of robot intelligence more than the sum of the accumulation of the past ten years of technology."

To some extent, this is a reflection of the current development of the mobile robotics industry.

In the past ten years, the mobile robot industry has emerged rapidly in China, and the mobile robots shuttling between logistics warehouses and factory production lines have become synonymous with intelligent logistics and intelligent manufacturing. According to GG Robotics data, from 2014 to 2023, the industry compound growth rate exceeded 40%, and the sales volume increased more than 20 times in ten years.

Under the potential space, in addition to startups, mature robot manufacturers and traditional logistics and automation equipment suppliers have entered the field, there are also a number of players across the border to kill, and the involution is serious. The other side of the coin is that after a decade of development, the industry scale in 2023, although more than 120,000 units, but the scale and standardized delivery is still a common industry to cross the difficult problem.

Will this track usher in a new story with the addition of big models? How should the industry get out of the involution narrative?

01 Some crowded track

Recently, service robotics company Purdue released its first robot PUDU T300 for industrial scenarios, entering the low-frequency, lightweight industrial delivery segment from catering and other service scenarios.

Zhang Tao, founder and CEO of Purdue Robotics, said in an interview that the reasons for entering the industrial market include the high potential of the industrial scene and the high acceptance of products such as robots, while they have previously accumulated experience in the field of service robots. The outside world interprets Purdue's cross-border entry into the industrial logistics and transportation

segment as an expectation to find a new growth channel as the industry rolls on.

However, observers have also pointed out that Purdue is actually entering a track that is already somewhat crowded.

Over the past decade, the domestic mobile robot industry has experienced the process of starting to catch up with overseas enterprises, to gradually improve the industrial chain, and the current localization rate has exceeded 90%.

An industry veteran introduced that the track began to ignite from March 2012, when Amazon acquired warehouse robot provider Kiva Systems for $775 million, which successively attracted a large number of enterprises to layout the field of logistics and warehouse automation.

After that, under the multiple backgrounds of rising labor costs, recruitment difficulties and industrial upgrading in China, mobile robot products have gradually been applied in many scenarios such as retail and manufacturing.

In this process, a large number of players have entered the game. According to the data of CRM Industry Alliance 2023, there are more than 600 enterprises related to the domestic AGV/AMR industry chain, among which there are more than 220 propriety enterprises whose main products are mobile robots (AGV/AMR) for industrial applications. This number is more than that of both Europe and North America regions where mobile robots were developed earlier.

For example, various enterprises established in the first wave of mobile robotics startup boom (2014 to 2017), such as Jiji Jia, Haikang Robotics, Quick Warehouse, and Hai Zuo, have already emerged and developed towards scale, among which Jiji Jia and Hai Zuo have brightened their sales performance in the overseas market, and their layouts have been gradually perfected. However, by 2023, there are still startups joining this track and obtaining financing.

Some old robotics companies Kuka, Yaskawa and Fanuc have also extended their business from robotic arms to the AMR field, and developed a logistics and handling robotics business, typically such as Kuka Robotics, which set up an independent mobile robotics division in Shanghai in 2021 and released related products in the same year.

Players in the AI field have also entered the field. For example, Kuangyi Technology, started with AI algorithms, and after entering the security market through the integration of hardware and software, since 2017, the field of intelligent logistics has been viewed by Kuangyi as a new breakthrough point for AI to land in the industry scene.

The platform enterprises in the logistics industry are from the perspective of their own efficiency, investing in related enterprises or launching self-research products. In July last year, the logistics platform cargo Lala had invested in Tusker Robotics, betting on the robot market. The invested company, Tasco, is itself an intelligent handling robot developer founded in 2021, and their main product is an intelligent pallet robot.

E-commerce logistics platform Cainiao, on the other hand, in increasing automation investment in warehouse transportation and distribution and other links, has launched self-developed warehousing and transportation mobile robots, such as the self-developed pallet four-way shuttle announced by Cainiao to the public in 2022.

As a result, this track has gathered startups, mature robot manufacturers and traditional logistics and automation equipment suppliers, and even companies like Foxconn and ZTE.

Players gathered, but veterans believe that the market is still early, a manifestation of the industry scale. CMR industry alliance data show that in 2023 China's mobile robot (AGV/AMR) sales scale of about 21.2 billion yuan, sales of about 125,000 units. Even with the difference in the size of the statistical caliber, the industry market is still to be further opened.

The market is also inevitably a situation of involutional competition. Xinsong Robotics mobile robotics BG president previously specifically pointed out that in the market demand for power battery and photovoltaic industry projects, individual project sales of hundreds of millions of dollars, but most of the projects are the lowest price to win the bid as the basic, "the bid price is not the lowest, only lower".
Haikang Robotics prospectus submitted last year shows that from 2020 to 2022 mobile robotics business gross margin has a downward trend, and mentioned, "the intensification of competition in the industry has led to a certain magnitude of price reductions in the products during the reporting period, especially in the mobile robotics business is more pronounced".

02 Customized demand vs. scale growth

Although the competition is fierce, the market demand is still strong. A senior person introduced, in recent years, with the new energy vehicles, photovoltaic and power battery industry demand increases, mobile robots are accelerating the application of these industries, these several scenes also accounted for the bulk of the industry increment.

The first tier of players in the market has gradually come out of the scale growth trend. Haikang Robotics recently announced that AMR all categories of 100,000 units of products off the line, to some extent this is the industry's head players after ten years of development to deliver an answer sheet.

And behind this data also hides an industry problem: how do mobile robot manufacturers achieve scale development?

Mobile robot industry alliance statistics, in 2022, China's industrial applications mobile robot market, there are four companies business orders across the 1 billion threshold, sales of more than 100 million yuan of enterprises about 42, of which, 1 to 300 million yuan of enterprises have 28, a large number of players in the industry's revenue scale is difficult to break through the 300 million yuan mark.
The reason for this situation lies in the complexity of the enterprise demand scene.

Warehousing, logistics and industrial production scenarios vary widely in demand. Haikang robotics vice president Wu Yonghai mentioned in a conference speech, they face thousands of industries, even if the same industry, the same scene, the operation mode of different enterprises is completely different.

At the same time, in the complex application of the scene, the enterprise's goal priorities may also be different, which makes it difficult to achieve large-scale batch delivery in the industry as in other consumer electronics industries.

Tang Wenbin introduced, some industries such as cold storage, the demand side will focus on the degree of unmanned, hope that a high degree of unmanned, the warehouse can use less people. Some customers will pursue easy maintenance, less types of equipment, maintenance will be simpler. There are also companies will pursue the size of the storage and order out capacity in the space.
With all these factors in play, customization is inevitable to meet customer needs.

How to achieve a balance between the demand for customization and the development of enterprise scale? Mobile robot vendors are also trying to find ways to solve the problem, such as using componentization and platformization to meet the challenge. To sort out the complicated business functions, categorize and sample some atomic-level components, and then carry out atomic-level combinations to build different application scenarios to adapt to a variety of needs.

In addition, companies are increasingly emphasizing the ability of secondary modification and development at the customer site. Haikang Robotics mentioned that they have a specialized Dataflow business development platform, which can be choreographed with business logic, and users can continuously adjust the logic strategy according to the changes in their own business.

A senior person cited an example, "After the system, the customer found that there is a waste to change how to do? Can not change their own to find manufacturers, but manufacturers have retired, this time it is very embarrassing." The construction of the secondary development platform will be able to adapt to the changes in the demand side of the enterprise.

Software systems are also very important. One is the compatibility of different software products. Due to the complex needs of the enterprise business site, some solutions require different brands of products to cooperate with each other. But how to schedule the products of different companies to achieve compatibility, is also a problem. Another point is that, with some intelligent warehousing scenarios landing in depth, how to coordinate between large-scale mobile robots, robots how not to crash and plan the optimal route, very test the ability of the algorithm.

At present, the mainstream domestic logistics robotics enterprises are very important to the investment in software systems. For example, when Kuangyi developed the Hetu system in 2019, it was to use Hetu as a central brain to dispatch all the self-developed and three-party equipment, and realized the connection of software systems including the warehouse management system (WMS), the warehouse execution system (WES), the task execution system (TES) and the robot body control software. It not only realizes intelligent integrated management of warehousing and logistics, but also supports simulation at the planning stage, truly realizing the full life cycle management of warehousing and logistics system.

In the context of the industry's multi-product cooperation, to promote the relevant interface standardization and unification has also become a consensus within the industry. Some industry associations and have been in the lead to mention the industry standard, there are also enterprises have acted. For example, 2022 mobile robot industry alliance developed "industrial applications mobile robot and its scheduling system data interface specification" has been released in April last year.

At present, mobile robot companies need to face a systematic project. Tang Wenbin believes that the entire chain, from hardware, software, algorithms, to the entire program design and final implementation of the landing, including some details, such as customer interfaces, training and operation and maintenance, the entire long chain of business affects the final effect. "This is an integrated system, vendors can not be too short in any one link short board."

03 The Big Model Dividend

Over the past year or so, big model technology has triggered a chain reaction around the world and opened up a lot of room for growth.
IDC recently released a data, from October 1, 2022 to March 15, 2024 nearly 18 months, the world, including cloud computing vendors, hardware chips, to software applications and big model companies, 60 companies, in the big model wave, the market value of the capital market has increased by $ 8 trillion.

The arrival of big models has also brought a high level of discussion heat to the robotics track. A veteran of a robotics startup told Digital Intelligence Frontier that they've seen big models improve the level of robot intelligence more in the past year or two than the cumulative sum of the last decade's technology.

"Originally, it was based on writing dead programs and programming out the robot's intelligence." The source told Digital Intelligence Frontier that, for example, to get a robot to go from point a to point b to point c, the process needs to be clearly defined before the robot can go over. After the big model is added, as long as the user says where to go in natural language, the robot can realize it.

Human-robot interaction is changing. Tang Wenbin, CTO of Kuangyi Technology, gave an example, for example, in some European logistics and industrial production sites, there are a number of voice-based control system solutions, many people work with a headset, through the restricted instructions, using voice to interact with the system. After the arrival of the big model, the powerful semantic understanding ability, people can be faster and more convenient to give the operating instructions to the system.

Tang Wenbin told Digital Intelligence Frontier that another more important change is that the end-to-end large model, from perception, decision-making and control capabilities concentrated in a model, the ability of people to control the robot, the machine's degree of intelligence is expected to be on a new level.

In fact, the end-to-end model is also the current industry's basic big model evolution and development direction, such as OpenAI recently released GPT-4o is the end-to-end model, which has been realized as the input of any combination of text, audio and image, and to generate any combination of text, audio and image outputs, the interaction response speed and real-life dialogue is very similar. The delay effect released by Google in August last year has amazed many people Robotics Transformer 2 (RT-2 for short), a large model of robotics, is also an end-to-end model that integrates vision-language-action (VLA) capabilities.

The person in charge of an industry solution of a domestic artificial intelligence enterprise told Digital Intelligence Frontier that the end-to-end big model means that from the input of speech to semantic understanding to control decision-making, which was previously done by multiple models stacked on top of each other, is now done by one model. Previously, the superposition of the way more or less will bring the loss of the degree of intelligence, while the flow of information between different models also naturally have more delay, while the end-to-end model means less loss, a higher degree of intelligence and shorter latency.

Some people have questions, logistics and warehousing link looks relatively closed, this scene is necessary to introduce a large model? Tang Wenbin believes that warehousing logistics scenarios in fact, the type of task is not a single. For example, a material transportation operation scene contains moving boxes, picking, packaging, taping and other types of tasks, the generalization ability of the big model can bring a very definite value. "The logistics scenario is, in fact, a very good one. It has quite a lot of work to do and needs smarter robots."

These new changes are exciting players in the logistics robotics track. After the big model wave came last year, Tang Wenbin and his team quickly reacted, "Kuangyi Technology wanted to do robotics when it was founded, and with this wave, we are one of the few companies that understands both big models and robotics, and these two skill points rarely converge on a team", and they are currently building a combination of big models and logistics robots products.

Industry sources believe that the enhancement of the level of equipment intelligence will accelerate the application of products in all types of industry scenarios penetration, which may release new space to avoid players to roll in the stock space.

Haikang Robotics CEO Jia Yonghua has given an example. Previously, the traditional AGV perception ability and autonomous decision-making ability is relatively weak, can only realize the point-to-point handling or some of the loop line simple fixed repetitive work, the scale of the industry is also relatively small. He once went to a well-known company in Japan, the annual shipments of only a few hundred units. But as mobile robots to join more algorithms, monomer intelligence is more and more powerful, the product has become more flexible, can run in the human-computer interaction environment, the scope of application of mobile robots to achieve a large expansion, such as last year's year Haikang robots shipments may be more than ten years ago, the entire industry.

However, the dramatic changes brought about by large models may have just begun. The industry also generally believes that using Transformer models and pre-training to do robot control opens up new directions in technology, but embodied intelligence development is still in its early days. "There is still a need to keep improving its success rate as the data loops, and to really close the loop in some scenarios to generate business value. The road is still quite long." Tang Wenbin said.