New Knowledge from Sky Eye | Robots come from the future and collide with large models to create sparks of 'wisdom'

Foreword: With the in-depth development of China's digital economy, the process of industrial digitalization continues to accelerate, and big data plays an increasingly important role in industrial decision-making. Tianyancha Data Research Institute hereby launches the "Tianyan New Knowledge" column, which relies on Tianyancha big data to sort out the industrial structure and development context, interpret the latest trends in the industry and investment and financing trends, and provide reference for decision-making by all parties.

This article is the thirty-seventh article in the Tianyan New Knowledge·Industrial Analysis Series.

Robots, the first "close contact" of most people comes from science fiction movies and from the endless imagination of transcending real life. Robots gradually evolve from "tools" to "intimate partners".

The "big model" that became popular at the beginning of the year, and the humanoid robot that became "popular" again, made people once again exclaim about the changes in lifestyles that technology has had.

In the future, with the continuous maturity of AI technology and breakthroughs in the research and development of humanoid robots, the "vision of robots entering the lives of ordinary people" will become closer and closer to reality.

1. Robots: Evolving from “automation” to “intelligence”

The International Federation of Robotics (IFR) proposes that a robot is a drive mechanism that can be programmed on two or more axes, has a certain degree of autonomy, and moves in a specific environment to perform predetermined tasks.

1. Robot application scenarios gradually move from "simple and elementary" to "complex and advanced"

Initially, the core function of robots was to replace people in repetitive and dangerous work while improving efficiency and accuracy; later, robots with the function of "serving people" came into people's eyes and were used for receptions and other purposes away from humans. More recently, entertainment/sweeping robots have entered homes on a large scale; then their high-precision features have been used in logistics and medical care, and automatic pickup and auxiliary care robots have begun to appear.

2. Robots gradually evolve from "automation" to "intelligence"

The development of robots has gone through three generations. The first generation is a program-controlled robot: action instructions are input into the robot through programming or teaching. However, due to the lack of external sensors, the robot can only complete the actions prescribed by the program. Once the environmental conditions change slightly, problems will occur in the robot's work; the second generation is the adaptive robot: it has vision, force sense and other sensors, and can adjust the working state based on the information obtained by the sensor; the third generation is the intelligent robot: its With a richer set of sensors, it can not only acquire and process external comprehensive information, but can even set its own action goals accordingly. Its intelligence is mainly reflected in three aspects: perception interaction, independent decision-making, and self-optimization.

2. Large model: First-tier cities take the lead in developing general artificial intelligence

The April 2023 meeting of the Political Bureau of the CPC Central Committee pointed out that we must attach great importance to the development of general artificial intelligence, create an innovation ecosystem, and pay attention to risk prevention. Different from the "artificial intelligence" generally referred to in previous important meetings such as the Central Economic Work Conference, this Politburo meeting clearly emphasized "universal", that is, the universal AI large model technology route led by Chat-GPT since November last year.

The release of Beijing’s “Several Measures” is the first special measure proposed by a local government in China to closely follow the industrial development of AI large models. Beijing has fired the first shot of the local large model competition. The current "thousand-model war" triggered by Chat-GPT has begun. Large models may bring a new efficiency revolution and experience improvement to different industries. As Beijing, the first city of AI, takes action, Shanghai, Shenzhen, Hangzhou and other regions may also take action one after another.

3. Humanoid Robot: The "unexpected encounter" between man and machine

Humanoid robot is an intelligent robot whose appearance and movement methods are similar to humans, and its definition and purpose of use are also like humans. Humanoid robots, also translated as "humanoid robots", literally mean robots designed and manufactured to imitate human form and behavior. There is currently no universal definition of humanoid robots, but according to the professional book "Humanoid Robots", humanoid robots should be able to "work in environments where humans work and live, operate tools and equipment designed for humans, and communicate with humans." Under this premise, humanoid robots should eventually have a human-like body structure, including head, torso, and limbs, use bipedal walking, use multi-fingered hands to perform various operations, and have a certain degree of cognitive and decision-making intelligence.

Humanoid robots started in the late 1960s, with Japanese research results attracting the most attention. In 1973, Professor Kato Ichiro of Japan's Waseda University developed the world's first humanoid robot, the WL-5 biped walking machine WABOT-1. Strictly speaking, it is a bionic machine and is the prototype of a humanoid robot. Honda began researching humanoid robot ASIMO in 1986 and successfully released the first prototype in 2000.

4. Humanoid robots and AI large models: universal scenarios accelerate the revolutionary advancement of the C-side

With the continuous breakthroughs in key technologies such as integrated design technology, motion management and control technology, and sensor perception technology, as well as the continued integration and application of new generation information technologies such as artificial intelligence and 5G, special robots are being used in coal mines, deep seas, and polar regions at an accelerated pace and other scenarios, releasing huge production and scientific research value. Among them, what fascinates cutting-edge technology companies and ordinary consumers the most is the emergence and iteration of intelligent mobile robots represented by humanoid robots.

At present, AI technology makes it possible for robots to operate autonomously by building an intelligent system with comprehensive perception, real-time interconnection, analysis and decision-making, and autonomous learning. AI enhances the robot's perception capabilities through robot vision technology, and improves its ability to analyze, make decisions, and learn independently by building algorithm models, so that the robot can complete tasks independently.

1. The ability to perceive the world (robot eyes)

Laser and visual navigation are the mainstream application solutions in the perception and positioning technology of robot autonomous movement. The development of computer vision has gone through traditional vision methods represented by feature descriptors and deep learning technology represented by CNN convolutional neural networks. Currently, general visual large models are in the research and exploration stage, and the scenes of humanoid robots are more complex than industrial robots. AllinOne's multi-task training solution, which is versatile, more complex, and has large visual models, can enable robots to better adapt to human life scenarios.

On the one hand, the strong fitting ability of large models enables humanoid robots to achieve higher accuracy in tasks such as target recognition, obstacle avoidance, three-dimensional reconstruction, and semantic segmentation; on the other hand, large models solve the problem of deep learning technology Over-reliance on single task data distribution and poor scene generalization results. General vision large models learn more general knowledge through large amounts of data and migrate it to downstream tasks. Pre-trained models based on massive data have better performance Knowledge completeness improves scene generalization effect.

Typical product: Tesla "Optimus (Optimus Prime)"

At the perceptual level, the Tesla robot head uses 8 cameras to collect visual information. At the computing level, the robot will use the FSD (Full Self-Driving, fully autonomous driving) computer currently used in Tesla cars, and use models such as neural networks to process information in real time. Tesla plans to use the supercomputer "Dojo" to train the artificial intelligence model used by the robot to improve its efficiency in identifying and responding to external objects.

2. The ability to think and make decisions (robot brain)

The current robots are all special-purpose robots and can only be used in limited scenarios. Even robot grabbing, based on computer vision, is still in a limited scenario. The algorithm is only used to identify objects. How to do it and what to do still need to be done. Definition of person. If you want to make a robot universal and ask it to water the flowers, it will know how to get the kettle, pick up the water, and then water the flowers. This is something that requires common sense to complete. Until large-scale models are available, it will be nearly impossible to equip robots with common sense. Large models allow robots to possess common sense and thus have the versatility to complete various tasks, completely changing the model of universal robot implementation. The adaptability of human tools to the environment eliminates the need to build tools for robots.

Typical product: The first robot citizen "Sophia"

In 2017, Sophia became the first robot in the world to receive citizenship. She said she would use her wisdom to help humanity develop and let us not be afraid of her, she was very friendly. In 2018, she also became the first AI teacher in the world to conduct online education courses. Sophia said that future robots are fully qualified for the work of teachers and can effectively solve the psychological and emotional problems encountered by students based on interaction with students.

3. Execution ability (robot limbs)

Mobility (legs) and fine manipulation (hands). Making robots into humanoid shapes is to make the robot's execution capabilities more versatile. The environment in which robots perform tasks is built according to human body shapes: buildings, roads, facilities, tools, etc. This world is designed to facilitate humanoid creatures like humans. People must redesign an environment suitable for new robots to cope with their emergence. It is relatively easy to design a robot to perform tasks within a specific range. If you want to improve the versatility of the robot, you must choose a humanoid robot that can serve as a clone. Additionally, humanoid robots are easier to communicate with emotionally than humans because of their ability to make people feel close.

Typical product: Boston Dynamics Atlas

In December 2020, Boston Dynamics released a video of Atlas dancing, with smooth and expressive movements. The robot must adjust its posture while jumping in the air to maintain balance and accurately complete dance movements. In August 2021, in the latest official video, Atlas can "parkour" in obstacle environments and perform a series of difficult full-body movements such as jumping, diving, rolling, and somersaults.

4. “Embodied Intelligence” Robot: The Ultimate Form of Artificial Intelligence

The concept of "embodied intelligence" originated from the problem of how to give computers the perception and action capabilities of a one-year-old child. It can be simply understood as a variety of different forms of robots, allowing them to perform various tasks in a real physical environment to complete the evolutionary process of artificial intelligence, such as: humanoid robots, intelligent driving cars, or future " Transformers".

Nvidia founder Huang Jenxun said at the ITFWorld 2023 Semiconductor Conference that Embodied AI is an intelligent system that can understand, reason, and interact with the physical world. It is the next wave of artificial intelligence.

The greatest characteristic of embodied intelligence is the ability to autonomously perceive the physical world from the perspective of the protagonist and use an anthropomorphic thinking path to learn, thereby giving behavioral feedback expected by humans, rather than passively waiting for data to be fed. By providing a learning and feedback system based on human behavior, humanoid robots provide an iterative foundation and experimental field for exploring more complex behavioral semantics. Therefore, the gradual improvement of humanoid robots also provides a direction for the implementation of embodied intelligence. It is an important application scenario of embodied intelligence and will also provide direction and space for the iterative optimization of embodied intelligence.

5. Patented technology: patent applications for robots and large models are increasing year by year

1. Robot patent:

In recent years, robots have become more closely integrated with cutting-edge technology industries, such as big data and artificial intelligence. Especially the emergence of humanoid robots has made robots as a technology industry increasingly more "technical".

From a quantitative perspective, robot patents have maintained steady growth, with more than 23,000 patent applications in 2022; from a type perspective, invention patents and utility models account for a high proportion, with the sum of the two accounting for 90% of the total.

2. Large model patent:

The "outbreak" of large models is not without signs. In the field of artificial intelligence, as its "crown jewel", natural language processing is a technology for understanding human language, emotions and thoughts. From its "few words" introduction, you can find out what its role in "machine" is. The importance and extremely high difficulty of "what is a human being".

Judging from the number of patent applications in recent years, it has maintained a relatively stable and rapid growth. Since 2018, the compound growth rate of its application volume has reached 61.2%;

In terms of types, invention patents occupy an absolute leading position, accounting for nearly 98%.

According to Tianyancha Research Institute, the use of large-scale model technology to promote artificial intelligence has achieved a historic leap, and there are still many broad innovation spaces to explore in the future. China's high-level economic and social development provides sufficient scenarios and data foundations for innovative large-scale models, so artificial intelligence has huge development potential in China.

The above is the detailed content of New Knowledge from Sky Eye | Robots come from the future and collide with large models to create sparks of 'wisdom'. For more information, please follow other related articles on the PHP Chinese website!