Tesla completely abandoned radar and bet on cameras

FSD ninth edition upgrade has been officially released.

Tesla completely abandoned radar and bet on cameras

On July 23, 2021, with the ninth version upgrade of Tesla’s Fully Automatic Driving (FSD), the system completely abandoned radar input and relied entirely on the car’s camera. The system has been in beta testing since October 2020. After its official release, it will be sold in the form of additional components at a price of US$10,000. 

In May 2021, Tesla announced that it would completely remove radar from Model 3 and Model Y cars made in the United States, and hinted that the company is redoubled efforts to develop a strategy that is inconsistent with most other autonomous driving projects. There are two main reasons for Tesla’s path to pure vision. The first point is that the cost of lidar is extremely high. 

The second point is that the entire automatic driving system is based on the way that humans recognize obstacles. Human drivers do not have signal input devices such as lidar to judge road conditions during driving. On the contrary, human drivers can only judge the road conditions with the naked eye. By means of reinforcement learning, Tesla believes that machines should have the same decision-making ability, and therefore bet on pure vision. 

As early as 2019, Tesla CEO Elon Musk bluntly stated on the first Tesla Autopilot Open Day that “Anyone who relies on lidar is doomed to fail.” Of course, he has made countless enemies because of this. 

Artificial intelligence relies on data. At present, Tesla has conducted road video training of about 1.5pb of data for its autopilot system. Andrej Karpathy, the company’s director of artificial intelligence, believes that the new pure vision solution is better than traditional lidar and vision. combination. 

In addition, Tesla built a supercomputer today, specifically designed to deal with complex road conditions. According to Tesla’s data, the computing power of this computer is ranked fifth in the world. 

The biggest selling point of lidar is to achieve incredibly accurate depth sensing by reflecting laser light from objects, but through methods such as color judgment and triangulation, vision-based systems can also estimate depth, shortening the cost of cameras and lidars. gap. In 2019, Tesla released a camera-based pseudo lidar solution that converts camera-based depth estimation into the same type of 3D point cloud used by lidar, thereby significantly improving accuracy. 

Looking at the camera and lidar from a digital perspective, there may be a gap of less than 1% between the two. For Tesla, when they find that vision can reach 99% of the effectiveness of lidar, they will naturally think that reaching 100% is also possible. However, this is also the core of the long tail problem. Few autonomous vehicles dared to bet on pure vision like Tesla. This is also a commonplace question. The industry generally believes that the combination of camera and lidar is the best solution. reliable. 

Posted by:CoinYuppie,Reprinted with attribution to:https://coinyuppie.com/tesla-completely-abandoned-radar-and-bet-on-cameras/
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.

Like (0)
Donate Buy me a coffee Buy me a coffee
Previous 2021-07-26 03:47
Next 2021-07-26 03:49

Related articles