Tesla announces anti-traditional autonomous driving technology. Are gaming graphics cards more reliable than lidar?

In addition to new energy, autonomous driving technology is also a project that many car companies are actively researching. As a "big boss" in the field of autonomous driving, Tesla recently announced an autonomous driving technology that breaks with traditional technical means. This technology will abandon the laser radar that is generally favored by the industry, use images as the source of autonomous driving information, and complete autonomous driving through AI processing of image information. Musk said that this autonomous driving system will be more than three times stronger than the one currently used by Tesla.

At Tesla's annual Autonomy Day, Musk said, "Any self-driving company that uses LiDAR is doomed to fail." Although this statement is too negative about LiDAR, Musk has his own understanding. In Musk's view, LiDAR has a very low cost-effectiveness. It is OK to use it on spacecraft, but it is a waste to use it on cars. Secondly, the range of information that Aurora Radar can collect is limited to the surroundings of the vehicle, and all of them are digital signals, which cannot achieve true "visualized" autonomous driving.

Therefore, image-driven autonomous driving technology has become the research direction of Tesla's new autonomous driving technology. In the view of Andrej Karpathy, director of Tesla's artificial intelligence and autonomous driving vision, "in a sense, LiDAR is a shortcut, but it avoids the basic problem of visual recognition that is very important for autonomous driving, and its collection of road information is not comprehensive enough." Therefore, "computer vision" that mainly collects information through cameras has become a new direction for Tesla to study new autonomous driving technology. Because vehicles, roads, markings, pedestrians, traffic lights, etc. can be captured by cameras, cameras are more intuitive and comprehensive than LiDAR in obtaining road information, and vehicles will also be safer and smarter in autonomous driving.

However, the amount of image data collected by the camera is much more than that of the laser radar, so the entire system has higher requirements for image processing chips. This is when NVIDIA comes into play, yes, the gaming graphics card in your computer. As the inventor of the graphics processor GPU, NVIDIA tailored a "graphics card" for Tesla to collect and process images taken by the on-board camera. This is NVIDIA's Tegra/DGX image hardware platform. This technology is not only used in Tesla's autonomous driving, but also in the processing of game images.

In addition to the technology provided by NVIDIA, Tesla also announced its self-developed chip, Tesla FSD. Tesla FSD is an FPGA chip manufactured using Samsung's 14nm FinFET process. It integrates 6 billion transistors and 250 million logic gates, 32MB SRAM cache, and a single processor can process 1TB of data per second, with a nominal performance of 36TOPS. A complete autonomous driving circuit board will integrate two Tesla FSD chips, executing a dual neural network processor redundant mode. The two processors are independent of each other, and even if one has a problem, the other can still work normally. Musk said that Tesla FSD is the most powerful chip in the world, and its performance supports an autonomous driving solution that is 21 times that of the current solution provided by NVIDIA. It is reported that Tesla FSD has been running on Model S and Model X for more than a month, and has been tested on Model 3 for more than 10 days.

Of course, this Tesla FSD chip is still in the testing phase, and Tesla's subsequent upgrades in autonomous driving for users will still be based on Nvidia's Tegra/DGX platform.

According to Cheyiquan, which ranks among the top ten in China, Tesla is truly leading the industry in the field of autonomous driving, whether it is the systems currently in use or the new technologies under development. In contrast, some domestic brands have exaggerated their autonomous driving with L2 (some are even lower than L2) level assisted driving, which is just a more advanced adaptive cruise technology, which is completely different from Tesla.