Why are prospective car companies eyeing the robot business?

Dancer imitating Tesla robot

Is this a marketing coincidence of players "not doing their job"? Or does it really hint at the future development direction of the industry? Self-driving cars have not yet really hit the road, but major car companies are trying to further expand their territory and begin to focus on the robot market.

On September 7, Xiaopeng Motors released its first intelligent robot horse "Little White Dragon".

On August 20, Tesla announced that it will launch a humanoid robot called "Optimus" next year.

On August 18, Baidu mentioned its future ideal self-driving concept car "Car Robot" for the first time.

On August 10, Xiaomi showcased its self-developed first-generation bionic robot CyberDog "Tiedan" at its new product launch conference and announced the establishment of the Xiaomi Robotics Laboratory.

Within a month, "AI robot", which is not a new thing, has become a keyword that appears frequently in the intelligent driving industry, triggering many speculations.

Is this a marketing coincidence of players "not doing their job"? Or does it really hint at the future development direction of the industry? What is the confidence of smart driving companies to cross over to make "AI robots"? If they enter the robot market, what type of robots will be more suitable?

1. Ready to make a move

At least Musk hinted at the possibility of Tesla making robots early on.

In an earnings call on April 27 this year, Musk said that although people now see Tesla as a car company or an energy company, in the long run, because Tesla has one of the most powerful hardware and AI software teams in the world, Tesla will also be an AI robotics company.

Tesla is indeed putting it into practice step by step.

"Because we already make robots with wheels (Tesla cars), we now have almost all the parts we need to make humanoid robots." According to the information announced by Musk on Tesla AI Day on August 20, Tesla's humanoid robot Tesla Bot prototype will be launched next year. It can do housework and repair cars. Its purpose is to replace humans in repetitive, dangerous and boring work.

According to official information, the Tesla Bot prototype looks similar to humans, is about 1.73 meters tall, weighs 56.7 kilograms, moves at a speed of 8 kilometers per hour, can lift a maximum of about 20 kilograms, and has a head screen for displaying information.

Specifically, Tesla Bot is equipped with an FSD computer and an Autopilot camera , has a total of 40 joints throughout its body, and has force feedback sensors on both feet. Almost all sensors in Tesla cars will also be used in this robot.

At the end of August, Tesla also began recruiting robotics experts related to Tesla Bot, announcing 12 new positions including mechanical engineers, senior humanoid mechatronic robot architects, senior humanoid modeling robot architects, and embedded firmware engineers.

Unlike Tesla's humanoid robots, Baidu's conception of its robots is more based on existing self-driving cars and can be considered a wheeled robot in a broader sense.

According to Robin Li's vision, the ideal autonomous driving concept car "car robot" in the future will not have a steering wheel, accelerator or brake, but will be replaced by a huge curved screen and smart console.

In addition to having L5 level autonomous driving capabilities, the car robot also has interactive capabilities such as voice and face recognition. In addition, it should be able to continuously learn and upgrade to provide driving and services that meet passenger preferences.

"Car robots are different from what everyone imagines as cars. In fact, they will be more like robots, and this will also be the direction of car evolution in the future," said Robin Li at a previous press conference.

Compared with the "technical" robot concepts of Tesla and Baidu, the four-legged robots "Tiedan" and "Little White Dragon" of Xiaomi and Xiaopeng are much more down-to-earth. For example, "Little White Dragon" is designed to be rideable and positioned as the first smart transportation tool for children, while "Tiedan" has lowered the price of the four-legged robot to less than 10,000 yuan.

Xiaomi's "Iron Egg"

It is understood that the motor technology that controls the movement of "Tiedan"'s calves comes from Xiaomi's Mijia vacuum cleaner, and the core algorithm is based on the open source MIT Mini Cheetah and the open source robot operating system ROS 2. "Little White Dragon" integrates Xiaopeng Motors' self-developed intelligent driving technology. In order to explore more functions of "Tiedan", Xiaomi will also set up the Xiaomi Robot Laboratory and establish a "Robot Open Source Community". Xiaopeng is recruiting a large number of robot-related talents through the Pengxing Intelligent Team.

Traditional car companies are certainly not absent, and some have even entered the market earlier. After all, ASIMO, the world's first truly bipedal robot, was developed by Japan's Honda.

Among today's traditional car companies, Hyundai Motor is the most aggressive in deploying its robotics business.

In June this year, Hyundai Motor announced the completion of its acquisition of Boston Dynamics, a bionic robot research and development company. Prior to that, at the end of September 2020, Hyundai Motor also established a new studio (New Horizons Studio) focusing on Ultimate Mobility Vehicles (UMV), hoping to redefine vehicles through the combination of robotics technology and wheeled motion technology.

Hyundai's concept car robot "Elevate"

2. What are the common points?

So what do robots and autonomous driving technologies have in common? Or, why are car companies like Tesla and Xiaopeng capable of making robots?

In fact, the three key technologies common to self-driving cars and robots are: environmental perception, planning and decision-making, and execution control.

Lu Zhangyuan, director of Gaogong Robot Industry Research Institute, told Xin Zhijia that self-driving cars can be seen as another form of robots. The difference is that robots focus more on human-like attributes, while cars focus more on vehicle attributes.

"In fact, driverless cars are also a form of robots, but driverless cars imitate human driving behavior, while robots imitate humans in all aspects, including body movements, interactive behaviors, learning abilities, etc." Pang Jianxin, vice president of UBTECH and deputy director of the research institute, pointed out to New Intelligent Driving.

In recent years, Toyota has been focusing on researching the application of artificial intelligence technology in automobiles and robots, and has also released a number of bionic robots.

Gill Pratt, Toyota's chief scientist and CEO of Toyota Research Institute, said in a previous interview with the media that outdoor cars actually do very similar work to indoor robots. They both move people and objects from one place to another to meet people's life needs.

"The level of automation required for different purposes varies, but the required sensing, perception and decision-making technologies are almost the same. Therefore, we believe that the results of research on artificial intelligence will have very wide applications in both automobiles and household robots ," said Gill Pratt.

For example, in 2017, when most sweeping robots were unable to build accurate indoor maps and were not growing as rapidly as they are today, 360's autonomous driving team launched a sweeping robot based on the Lidar-Slam  laser radar , ultrasound , odometer, and gyroscope / accelerometer in the field of unmanned driving, which achieved map drawing, environmental positioning, obstacle perception and avoidance, route planning and control.

In addition, in 2018, the MIT Bionic Robotics Laboratory gradually opened up the motor driver and code of its robot dog Cheetah Mini, and the existence of the ROS open source robot operating system also made the research and development of robots modular, greatly lowering the threshold for robot software development.

In fact, Baidu's Apollo also uses the ROS operating system, but has made corresponding customized modifications based on the needs of autonomous driving.

In addition, on the hardware platform, self-driving cars and robots can also be basically compatible.

For example, NVIDIA's autonomous driving processor platform (NVIDIA DRIVE) can simultaneously meet the high performance and high computing power required by autonomous vehicles and robots.

However, there are also obvious differences between robotics and autonomous driving technologies. Overall, robotics is about both software and hardware, while autonomous driving companies may have stronger advantages in software.

Pang Jianxin said that this is because driverless cars and robots are used in different scenarios and the problems they need to solve are also different.

In most cases, driverless cars are driving on the road in structured scenarios, and the decisions they make must comply with established traffic rules. For safety reasons, the perception system of driverless cars needs to be foolproof.

The vast majority of robots' working scenarios are unstructured, with no rules or restrictions, and obstacles are mostly dynamic, which makes them more complicated in comparison.

"Due to structural features, cost and other reasons, the perception capabilities of most robots are weaker than those of driverless cars," said Pang Jianxin.

The technical challenge for smart driving companies that want to make bionic robots is that although the control links of vehicles are relatively mature, modern cars generally use Ackermann steering because they only have four wheels. Compared with robots that focus more on mobile robustness in unstructured environments and have dozens of joints, current smart driving companies still need to spend more time on motion control and corresponding engineering capabilities.

Tesla has also seen this, and most of the Tesla Bot-related positions that Tesla is recruiting for mentioned above are also related to this.