The humanoid robot that drives Musk crazy! Trillion-level new blue ocean, revealing the industry behind

Welcome to the WeChat subscription number of “Sina Technology”: techsina

Source: Zhixing (ID:zhidxcom)

Humanoid robot, the ultimate form of AI.

On June 3, local time, Tesla CEO Elon Musk announced on Twitter that a prototype will be released at Tesla Artificial Intelligence Day on September 30. It’s called Optimus, also known as the Tesla Bot, and it’s Tesla’s most important product this year.

The emergence of humanoid robots can empower thousands of industries and is the next wave of artificial intelligence scenarios. With the continuous maturity and commercialization of technology, it is expected to bring an unprecedented blue ocean of trillions.

In this issue’s intelligent internal reference, we recommend Huaxi Securities’ report “Tesla Bot: The Sea of ​​Stars in AI” to analyze the market prospects of humanoid robots. If you want to collect the report of this article, you can get it by replying to the keyword “nc634” in Zhishi (public account: zhidxcom).

01.

humanoid robot

A new chapter in AI

Tesla may launch its first humanoid robot prototype on September 30, 2022, and name it “OPTIMUS”. As early as August 19, 2021, Musk proposed the launch of a humanoid robot at the Tesla Artificial Intelligence Day, with the intention of solving dangerous errands that are highly repetitive and monotonous.

Musk’s announcement to enter the field of AI robots means that Tesla is not just an electric car company, but an AI company. In addition, Musk claims that Tesla robots will one day be more important than car companies over time.

The Tesla robot can simply be split into two domains, namely the AI ​​domain and the technical domain.

AI domain: FSD computer is used as the computing power core, equipped with 8 Autopliot Cameras as sensors, which supports deep learning, data analysis, Dojo training, automatic labeling and other algorithms.

Technical field: The head of the robot contains an information screen to display information. In addition, the robot is composed of lightweight materials, and the limbs contain about 40 electromechanical actuators, and the force feedback induction system is used to achieve smooth and agile walking on both feet.

According to Musk, the robot is about 1.73 meters and weighs about 56.7 kilograms. It can pick up about 20.4 kilograms of goods and walk at a maximum speed of about 8KM/hour.

The AI ​​domain is the core of humanoid robots, because robots can only complete specified tasks through continuous machine learning training. In addition, the Tesla humanoid robot is the master of Tesla’s automatic driving. Because the core of the humanoid robot shares the FSD system with intelligent driving, we expect that many neural network systems for intelligent driving will be applied to humanoid robots.

Data is the foundation for realizing intelligent driving and intelligent robots, and computing power provides the basic power for machine learning and neural networks. With the exponential growth of data processed by Tesla, the company abandoned the Nvidia A100 GPU as a supercomputer due to power consumption problems. Instead, relying on its own strong vertical integration capabilities, it developed the Dojo D1 chip focused on deep learning training, so the Tesla Dojo supercomputer came into being.

1. Brain: D1 chip

As the key unit of the Dojo supercomputer, the D1 chip achieves super computing power and super bandwidth, and achieves a balance between space and time. The chip adopts a distributed structure and a 7-nanometer process, equipped with 50 billion transistors, 354 training nodes, and the internal circuit alone is 17.7 kilometers long.

The Dojo supercomputer is a “performance beast”, with a computing power of up to s 9PFLOPs. The training module of the Dojo supercomputer consists of 1,500 D1 chips, with a total of more than 530,000 training nodes. The delay between adjacent chips is low. With Tesla’s own high-bandwidth and low-latency connectors, the computing power is as high as 9PFLOPs. The world’s premier supercomputer. Compared with the industry, the performance at the same cost can be increased by 4 times, the performance at the same energy consumption can be increased by 1.3 times, and the space ratio can be saved by five times.

The Tesla Dojo D1 chip can be mainly disassembled into 4 parts, namely CPU, Switch, Mat mult, and SIMD.

The CPU, the central processing unit, is the operation and control core of the computer series, and the final instruction unit for information processing and program operation.

Switch is a switch, which is a bridge between computer chips and chips, and has the function of data transmission.

SIMD is a single instruction stream with multiple data streams, which can be understood as parallel computing. It is a technology that uses one controller to control multiple processors and intervenes to achieve spatial parallelism. Simply put, one instruction can process multiple data.

Mat mult is a computing unit, which can focus on the calculation of the neural network, thereby accelerating the computing speed of the neural network, which is one of the fundamental reasons for Tesla’s computer to realize the beast of computing power. The computing unit can be understood as an artificial intelligence chip, namely an AI processor, which is a chip specially used for machine learning algorithms and neural network operations, which can be used for training and reasoning. Compared with the CPU and GPU of the same period, it can achieve 15-30 times performance improvement and 30-80 times efficiency (performance) improvement.

2. Soul: AI Machine Vision

Machine vision is an application and technical direction of AI deep learning. Whether it is humanoid robots or intelligent driving, it is one of the landing directions of machine vision.

Neural network is an important algorithm to realize AI deep learning, covering the whole process of humanoid robot from recognition to generating instructions. It is an artificial system with intelligent information processing functions such as learning, association, memory and pattern recognition, which is developed by modeling and linking the basic unit neurons of the human brain to explore a model that simulates the functions of the human brain system. Neural networks are widely used in intelligent robots, mainly in object recognition, planning, hypothesis, training//testing and other links.

The most important feature of neural networks is that they can learn from the environment and store the learning results distribution in the network’s synaptic connections. A neural network is a process of learning. Under the stimulation of the environment, input some sample patterns (input layers) to the network one after another, and adjust the weight matrix (hidden layer) of each layer of the network according to a certain learning algorithm. When the weights of each layer of the network converge to a certain value, the learning process will end (output layer).

Tesla has similarities in the path of intelligent driving and humanoid robots in machine vision. A complete set of training, testing ((working)) movements contains five parts: sensor, perception, evaluation, planning, and brake.

Tesla’s most famous AI algorithm is its pure vision solution in machine vision, which continues in the manufacture of humanoid robots.

Image-based object detection: The purpose is to determine whether there are object instances of a given class in the image, which can be dynamic or static objects, and if so, return the spatial location and coverage of each object instance. Object detection is the basis for solving more complex higher-level (temporal memory, etc.) vision tasks such as segmentation, scene understanding, object tracking, image description, event detection, and activity recognition.

Conversion from 2D object recognition to 3D object recognition: Tesla obtains the same object from different angles through cameras at 8 different positions, renders a 3D image of the object through a neural network (similar to the NeRF algorithm), and records the object Then generate a 3D vector space, through another neural network (similar to LSTM algorithm) and object recognition to calculate the position where the object will appear at the next time point, so far the humanoid robot has completed all the The perception step, which contains three-dimensional information and time dimension information, stores this information in the training set, and continuously strengthens the learning.

02.

bright future

ready

From smart cities to changes in the AI ​​wave of intelligent driving, it is expected that humanoid robots will be the next application scenario of artificial intelligence.

Big data era: In 2016, AI defeated Ke Jie. At the same time, with the improvement of basic computing power, my country has opened a new round of artificial intelligence boom, that is, the era of big data. Policies and capital come first, and application scenarios are gradually enriched. Drones, AI translation machines, etc. have landed one after another.

Intelligent driving: With the continuous explosion of massive data and the continuous evolution of basic computing power and chips, Tesla Autopilot has officially entered the world of intelligent driving with its perfect function definitions, algorithms that rely on continuous learning from data, and software upgrades through OTA. At the same time, Internet giants such as Google, Baidu, Tencent, and Huawei have entered the game one after another to promote the accelerated development of intelligent driving. Superimposed policies continue to promote the commercialization of autonomous driving. Today, domestic manufacturers in my country have made practical breakthroughs in smart cockpits and driving, and the localized ecology will be promising in the future.

Humanoid robots: In the future, with the implementation of humanoid robots, boring and repetitive tasks such as grocery shopping and housework will easily be replaced by humanoid robots. We believe that this is the next wave of artificial intelligence, and domestic companies are likely to copy the Achievements in the field of intelligent driving.

According to McKinsey data, with the continuous progress of AI, it is estimated that in 2030, about 375 million people in the world will be re-employed due to technological breakthroughs in AI. The global average replaced labor force ratio is 15%, and my country, as a country with a large population, is basically the same as the world at 16%.

In addition, according to Musk, the actual cost of humanoid robots will not be very high, and may be lower than cars. Andrew’s forecast is $25,000, or about 160,000 yuan. According to the minimum price of Tesla MODEL3 is about 280,000 yuan, a conservative estimate of the price of Optimus is 200,000 yuan. In the long run, it is conservatively estimated that by 2030, the global humanoid robot market will reach a scale of trillions, which is another unprecedented blue ocean of AI following the intelligent driving of trams.

The emergence of humanoid robots can empower thousands of industries and is expected to replace tedious, repetitive and tedious tasks. At the same time, a series of dangerous problems such as search and rescue are expected to be solved, and a series of scenarios such as express delivery, housekeeping, service industry, and industry are expected to be solved. It is expected to be the first to land. In addition, humanoid robots are the next wave of AI scenarios. With the continuous maturity of technology and commercialization, it is expected to bring an unprecedented blue ocean of trillions. Under the background of the gradual breakthrough and maturity of my country’s intelligent driving ecology, humanoid robots Robots are imperative, and domestic companies are likely to replicate the results achieved in the field of intelligent driving.

Manufacturers with self-developed AI processors can provide computing power support for the neural network of humanoid robots. The nature of artificial intelligence and the massive operations of data, compared to AI algorithms, data is the most important thing. The importance of computing power as the power source of data acceleration processing is self-evident.

According to the algorithm steps of machine learning, it can be divided into two parts: training and inference. The training process requires extremely large data input to support a complex neural network model. During the training process, due to the complex neural network structure and massive training data, the amount of calculation It is huge, so the computing power and efficiency (energy consumption) of the processor are extremely demanding.

The AI ​​processor chip can support the learning and accelerated computing of deep neural networks. Compared with GPU and CPU, it has double the performance improvement and extremely low power consumption level. In addition, the calculation volume of the inference link is relatively small compared to the training link, but it still involves a large number of matrix operations. Therefore, the artificial intelligence chip will play a large role.

The landing of humanoid robots requires data fusion in downstream scenarios, and manufacturers with AI algorithms have a comparative advantage. Tesla has achieved a pure vision solution in the field of intelligent driving, and the related FSD system can be directly used in the field of machine vision of humanoid robots.

However, before the commercial implementation of humanoid robots, its data needs to be closely integrated with downstream subdivision scenarios, and continuous iterative training is carried out through data and algorithms in high-quality subdivision scenarios, and finally valuable commercial services are provided. Humanoid robots cannot directly obtain massive data in subdivided scenarios, and companies that have commercialized AI algorithms have the advantage of being able to position themselves, that is, they are closely connected to downstream subdivision scenarios. The two parties can work together to empower customers and accelerate humanoid robots. The commercialization of robots has landed.

Zhishi believes that the goal of Tesla’s humanoid robot is to be applied in repetitive, boring and dangerous environments and working conditions, and will eventually go to our homes to completely solve the trend of continuous labor shortages. Although Tesla has a solid technical foundation in AI and robotics, whether Musk’s Haikou can come true may not be known until September.

(Disclaimer: This article only represents the author’s point of view and does not represent the position of Sina.com.)

This article is reproduced from: http://finance.sina.com.cn/tech/csj/2022-07-02/doc-imizirav1643709.shtml
This site is for inclusion only, and the copyright belongs to the original author.