How AI changes robots: Chen Runze’s new changes, new signs and new turns

Recently, the Chinese technology industry think tank "Jiazi Light Year" hosted "To You Who Chase the Wind and the Moon" 2 What needs to be rewritten is: 023 Jiazi Gravity Year-end Ceremony, in which, in the round table with the theme of "How AI Transforms Robots", Chen Runze, executive director of Source Code Capital, launched a sharing discussion around topics such as "new changes in the robot industry, new signs of AI transforming robots, and new investment directions".

He believes that new AI technology, especially large models, is changing the characteristics of robots. Now, robots have made significant progress in perception and understanding capabilities, which is a major change for the field of robotics and autonomous driving. Human-computer interaction has also made great progress, which is the area where large language models play the most direct role. In terms of interacting with the physical world, data and model scaling methodologies also show good potential, but the maturity of the technology needs to be carefully evaluated.

In terms of investment, he proposed two possible ideas. The first is to focus on cyclical opportunities for capital investment in downstream industries, and the second is to focus on the continued progress of general-purpose robots in terms of capabilities and costs

01

New changes: general intelligent robot

The rewritten content is: becoming a new topic that has attracted much attention this year

What needs to be rewritten in 2 for investment and financing in the robot industry is: 021 to 2 What needs to be rewritten is: 022 The first half of the year has cooled down. The reason may be due to some phased changes in the supply of capital and entrepreneurial assets. Variety.

What needs to be rewritten at 2 is: At the end of 2022, several events occurred that promoted changes in the industry. First, Tesla launched a humanoid robot, and secondly, teams such as Google successively showed many excellent large-scale models and robot-related demonstrations. These events drive industry growth. What needs to be rewritten before entering 2 is: In 2023, as the attention to general intelligent robots increases, more and more people begin to pay attention to the huge potential of reinforcement learning, large-scale models, diffusion models and other technologies in the field of robots, and the market Temperatures and moods also continue to rise. However, we should note that there may be some irrational factors

Looking back at the robot projects of the past few years, the main technical variables are positioning navigation and machine vision. Many mobile robots are relatively mature categories, and AI applications are mainly focused on object recognition, detection, and positioning and navigation. Many startups basically look for PMF in specific scenarios around these technologies. But we might as well broaden our horizons and look back at the outstanding companies that have developed throughout the automation industry in the past 10 years. We should not only focus on changes on the supply side, but also on downstream changes. Some relatively good and relatively large companies have been listed on the A-share market. In fact, these companies have strong industry attributes and have needs in the industry. There happens to be some robotics technology that can solve these needs. This kind of opportunity is the best. .

What needs to be rewritten is: 02

New signs: AI begins to revolutionize robots

In the past year or so, the progress of AI technology has given us greater confidence in robots. However,

customers actually don’t care whether you use AI or not. Taking some manufacturing scenarios as an example, customers only care about two issues: the first is whether it can meet performance requirements such as beat accuracy and throughput; the second is whether the cost is cost-effective. In addition to the cost of the robot itself, there are many other issues. A large part is the cost of delivery.

We have communicated with many robot entrepreneurs, researchers and engineers, and we clearly feel that the recent development of visual-language models has made a qualitative leap in the perception and understanding capabilities of robots. This progress will be used in the field of robotics and automation. Bringing rapid changes in the field of driving. When we put robots in an open environment, the robots are able to understand the semantics of the world, which is completely different from the way robots have been built in the past. In the past, machines' understanding of the world relied on the recognition of specific objects. For example, in the field of autonomous driving, the whitelist used by perception modules in the past was very limited, but today's semantic segmentation and understanding capabilities have been greatly improved, which has an important impact on subsequent planning. For another example, in the inspection scenario, detailed annotations were required in the past, but now the delivery cost of the inspection scenario may be significantly reduced

Regarding the problem of robots interacting with the physical world, we are closer than ever to finding a scalable and generalizable robot learning method. For specific tasks in some specific fields, we believe that technology has entered the engineering stage. But universal operating capabilities are still a luxury, at least in the short term. It has to be said that there have been many good studies on robot operating capabilities at the academic level recently. We have seen that robots in the laboratory can already handle rigid objects very well, and can even operate deformable objects such as clothes and plastic bags. But it takes time for these technologies to come out of the laboratory, and it requires the power of industry to push the flywheel forward.

With the development of large language models, human-computer interaction benefits more directly than physical world interaction. We will soon see opportunities for commercial applications. But it needs to be emphasized that when it comes to interacting with the physical world, one needs to carefully judge the maturity of the technology and its commercial viability

One of the keys to a universal robot is its ability to make autonomous decisions. We believe that the large language model has provided a good foundation for this, but the maturity of the technology needs further observation

We strongly feel that AI robots have reached the intersection of academia and industry. Source Code hopes to cooperate with entrepreneurs to find scenarios that can be started, roll up data, and explore reasonable forms of hardware in business. On this basis, promote the development of large-scale robot models.

03

Change in investment direction: focus on whether technology research and development can be applied in practice

When considering investing in a robotics company, you should pay attention to two main aspects

The first main line is to pay attention to the capital investment situation of the downstream industry in which the company is located, and whether the company can seize opportunities and combine them with its own technology. Some outstanding robotics and automation companies, such as Inovance, SUPCON, Maiwei, Pioneer Intelligence and Northern Huachuang, etc., all have distinct characteristics of capital investment in downstream industries. Before truly universal robots can be realized, we believe that robots must still be placed in the context of large-scale capital investment.

The second main line is universal. Source code is not eager to invest after the popularity of humanoid robots, but it also fully recognizes the need to find a hardware form with the greatest common denominator and reduce costs through mass production and manufacturing. On this basis, transforming robot application development into a job that is as close as possible to software development is a continuation of the path verified by devices such as PCs and mobile phones. Therefore, when evaluating the investment value of general-purpose robots, you need to pay attention to whether the company has done a lot of work around reliability, manufacturing, cost, etc. Of course, the convergence of any hardware form also depends on scale. Even for general-purpose robots, you need to find a suitable industry with certain capital investment. Therefore, the focus of the two main lines is actually demand.

The above is the detailed content of How AI changes robots: Chen Runze’s new changes, new signs and new turns. For more information, please follow other related articles on the PHP Chinese website!