Tesla has plans to launch its “Dojo” supercomputer in late 2021, according to a tweet by Elon Musk. Dojo will be used to support the neural network behind its Full Self-Driving software by processing and labeling the data it receives from vehicles’ cameras.
Version 1 is about a year away
— Elon Musk (@elonmusk) November 8, 2020
The data used to train the neural network is currently being processed by a combination of computer processing power and humans who can step in when the computer gets confused. Humans are still better than computers at recognizing objects in a world where the imagery would always be changing as Tesla vehicles drive along.
For instance, it may not be difficult to distinguish between road signs, but the neural network will also have to be able to tell the difference between a turn lane and a “straight” lane, recognize an exit lane (and also recognize that an exit exists and that’s not just a hill), and predict what other vehicles on the road are likely to do in an environment where it’s still normal to have a human at the wheel and they might suddenly cut a Tesla off.
Tesla may be banking on the idea that its supercomputer will help to improve the neural network’s ability to recognize objects in real-time as its Full Self-Driving software does its job. It will possess the raw ability to process petabytes’ worth of imagery generated by cameras on vehicles operating in the field.
Robotaxi Project Could Benefit
Dojo could be a major plus for Tesla’s plan to launch its fleet of Robotaxis as early as the first half of 2021. It’s been estimated that the Robotaxi could be worth billions of dollars in revenue every year as it cashes in on the “ride sharing” market.
Although some people might be leery about not having a human at the wheel when they call for a ride, autonomous driving backed by a supercomputer like Dojo could actually improve the safety of taxis and “ride share” vehicles by removing the risk of human error from the equation. Statistics show that human error could be a factor in up to 92% of traffic accidents.
Although some Teslas equipped with a prototype of Full Self-Driving have been involved in accidents, the Tesla was rarely at fault. Tesla’s AI currently possesses more than six billion miles’ worth of driving data, which is already enabling the addition of additional features like the ability to recognize speed bumps. This will mean not only a safer ride, but also a less jolting one for potential Robotaxi customers.
All this might call up images of cartoons in which the Jetsons could interact with their autonomous driving software or just trust it to drive them to their destination while they browse the news. The software does seem to have access to real-time traffic data and “road” conditions for their flying cars. It doesn’t always help them avoid traffic jams, but you also never see them get involved in a traffic accident.
Tesla may aim to do the same thing, though there is no word yet on whether it has a flying car in the works. Dojo is likely to help it improve the neural network by providing a large boost in processing power and the ability to recognize conditions on the road in real-time.