Tesla believes that machine vision will do better than radar and sensor detection in increasing the accuracy and speed of response of its car systems. At the CVPR 2021 workshop, Andrei Carpathi, senior director of Tesla’s AI unit, explained that the machine vision capability was to be brought to the automated driving system using a Tesla supercomputer called the Dojo.
According to Engadget, Karpati explained at the workshop that Tesla’s powerful supercomputer will be able to train artificial intelligence on a huge scale of data to operate at the same speed and accuracy as humans. The Dojo is the next generation of supercomputers with 1.8 exaflops processing power and 10 petabytes of NVM Express memory at 1.6 terabytes per second.
Not yet rated from Dojo; But Carparty believes the system will be one of the fastest in the world. “If you take into account the total number of flops, (the dojo) will be in fifth place,” he said. “The site is now owned by NVIDIA’s Celine supercomputer, which has the same architecture and number of GPUs.”
The Dojo supercomputer uses artificial intelligence from videos collected from eight cameras in Tesla’s car, each recording at 36 frames per second. This data volume seems difficult to manage; But they are more scalable than building and maintaining high-quality maps around the world.
This system has been able to work well in sparsely populated areas where cars can pass without encountering obstacles; But like other automotive manufacturers, Tesla has found that navigation in densely populated areas is much more difficult. However, Carpathian claims that Tesla Computer has been able to analyze data related to newer types of driving alerts, detect pedestrian collisions and incorrectly depress the accelerator pedal instead of the brake.
Tesla cars have crashed in the past due to obstacles or inability to navigate properly; But Ilan Mask, the company’s CEO, strongly supports the use of machine vision. He recently tweeted: “When radar and sight information are different, which one do you trust? Machine vision is much more accurate; “So it is better to strengthen the machine vision instead of combining the sensors.”
Tesla believes the supercomputer could eventually help cars achieve advanced automated driving capabilities; But it is better to wait to see the results of these experiments, because Tesla is accustomed to Ambitious promises Do not act on your own.