Tesla has showcased its massive supercomputer to train its self-driving system. As per the figures released, it could be the fifth most powerful supercomputer in the world, reports Electrek.
Tesla said that it’s using its new computing prowess to train the Autopilot feature on its vehicles. Besides, it will also train the unreleased self-driving artificial intelligence systems.
Way ahead in the self-driving race
It’s a step towards the company’s planned supercomputer called “Dojo,” a next-gen system capable of enabling Tesla vehicles to achieve an unmatched level of autonomy in real-life scenarios.
The mechanics of the system show how Tesla is committed to focusing on vision alone to train its vehicles to drive autonomously. This could also take them way ahead of competing systems that use radar technology as well.
“For us, computer vision is the bread and butter of what we do and what enables Autopilot,” said Karpahty, as quoted by Electrek. “And for that to work really well, we need to master the data from the fleet, and train massive neural nets and experiment a lot.”
“If you take the total number of FLOPS it would indeed place somewhere around the fifth spot,” Andrej Karpathy, Tesla senior director of AI told Engadget.
Elon Musk’s desire to go vision-only for Tesla
The team hasn’t run tests on the system to confirm its position on the TOP500 supercomputing ranking. The data provided to the computer is in the form of live video feeds from eight cameras mounted on a Tesla car. The system then processes this data, allowing the vehicles to have a better understanding of their physical surroundings.
The news succeeds Musk’s reiteration of his plans to go vision-only and not be dependent on radar data to make Tesla vehicles drive themselves. “When radar and vision disagree, which one do you believe?” Musk said in a tweet. “Vision has much more precision, so better to double down on vision than do sensor fusion.”
Speaking of supercomputers, Perlmutter, which is hailed as the world’s fastest computer for AI will be used to build the largest 3D map of the observable universe.
