It’s been nearly two years since Tesla’s first “Autonomy Day” event, at which CEO Elon Musk made numerous lofty predictions about the future of autonomous vehicles, including his infamous claim that the company would have “one million robotaxis on the road” by the end of 2020. And now it’s time for Part Deux.
This time, the event will be called “AI Day,” and according to Musk, the “sole goal” is to persuade experts in the field of robotics and artificial intelligence to come work at Tesla. The company is known for its high rate of turnover, the latest being Jerome Guillen, a key executive who worked at Tesla for 10 years before recently stepping down. Attracting and retaining talent, especially top tier names, has proven to be a challenge for the company.
The August 19th event is scheduled to start at 5PM PT / 8PM ET at Tesla’s headquarters in Palo Alto, California. According to an invitation obtained by Electrek, it will feature “a keynote by Elon, hardware and software demos from Tesla engineers, test rides in Model S Plaid, and more.” Much like Battery Day, the event will be livestreamed on Tesla’s website, giving investors and the media, as well as the company’s many fans, an up-close look at what’s under development.
Musk and other top officials at the company are expected to provide updates on the rollout of Tesla’s “Full Self-Driving” (FSD) beta version 9, which started reaching more customers this summer. We may also get details about Tesla’s “Dojo” supercomputer, the training of its neural network, and the production of its FSD computer chips. And there will also be “an inside look at what’s next for AI at Tesla beyond our vehicle fleet,” the invitation says.
Let’s start with what we know and work our way toward the speculation of what’s to come.
The big news out of Tesla’s first Autonomy Day was the introduction of the company’s first computer chip, a 260 square millimeter piece of silicon that Musk described as “the best chip in the world.” Originally, Musk had claimed that Tesla’s cars wouldn’t need any hardware updates, only software, on the road to full autonomy. Turns out that wasn’t exactly the case; they would need this new chip — two of them, actually — in order to eventually drive themselves.
A lot has happened between the 2019 event and now. Last month, Tesla began shipping over-the-air software updates for FSD beta v9, its long-awaited, definitely not autonomous, but certainly advanced driver assist system. That means that Tesla owners who have purchased the FSD option (which now costs $10,000) would finally be able to use many of Autopilot’s advanced driver-assist features on local, non-highway streets, including Navigate on Autopilot, Auto Lane Change, AutoPark, Summon, and Traffic Light and Stop Control.
The update doesn’t make Tesla’s cars fully autonomous, nor will it launch “a million self-driving cars” on the road, as Musk predicted. Tesla owners who have Full Self-Driving still need to pay attention to the road and keep their hands on the steering wheel. Some don’t, which can have tragic consequences.
Loved by fans, loathed by safety advocates, the FSD software has gotten Tesla in a lot of hot water recently. In recently publicized emails between Tesla and California’s Department of Motor Vehicles, the company’s director of Autopilot software made it clear that Musk’s comments (including his tweets) do not reflect the reality of what Tesla’s vehicles can actually do. And now Autopilot is under investigation by federal regulators who want to know why Teslas with Autopilot keep crashing into emergency vehicles.
Aside from the rollout of FSD beta v9, Tesla has also had to adjust to the global chip shortage. In a recent earnings call, Musk said that the company’s engineers had to rewrite some of their software in order to accommodate alternate computer chips. He also said that Tesla’s future growth will depend on a swift resolution to the global semiconductor shortage.
Tesla relies on chips to power everything from its airbags to the modules that control the vehicles’ seatbelts. It’s not clear whether the FSD chips, which are produced by Samsung, are being impacted by the shortage. Musk and his cohort may provide some insight into that during this week’s event.
Outside the car, Tesla uses a powerful supercomputer to train the AI software that then gets fed to its customers via over-the-air software updates. In 2019, Musk teased this “super powerful training computer,” which he referred to as “Dojo.”
“Tesla is developing a [neural net] training computer called Dojo to process truly vast amounts of video data,” he later tweeted. “It’s a beast!”
He also hinted at Dojo’s computing power, claiming it was capable of an exaFLOP, or one quintillion (1018) floating-point operations per second. That is an incredible amount of power. “To match what a one exaFLOP computer system can do in just one second,” NetworkWorld wrote last year, “you’d have to perform one calculation every second for 31,688,765,000 years.”
By way of comparison, chipmaker AMD and computer builder Cray are currently working with the US Department of Energy on the design of the world’s fastest supercomputer, with 1.5 exaFLOPs of processing power. Dubbed Frontier, AMD says the supercomputer will have as much processing power as the next 160 fastest supercomputers combined.
When completed, Dojo is expected to be among the most powerful supercomputers on the planet. But rather than performing advanced calculations in areas like nuclear and climate research, Tesla’s supercomputer is running a neural net for the purposes of training its AI software to power self-driving cars. Ultimately, Musk has said Tesla will make Dojo available to other companies that want to use it to train their neural networks.
Earlier this year, Andrej Karpathy, Tesla’s head of AI, gave a presentation at the 2021 Conference on Computer Vision and Pattern Recognition, during which he offered more details about Dojo and its neural network.
“For us, computer vision is the bread and butter of what we do and what enables Autopilot,” Karpathy said, according to 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. So we invested a lot into the compute.”
Earlier this month, Dennis Hong, founder of the Robotics and Mechanisms Laboratory at UCLA, tweeted a photo of a computer chip that many speculate is the in-house hardware used by Tesla’s Dojo.
But Hong is an interesting figure for other reasons, too. He specializes in humanoid robots and was a participant in the DARPA Urban Challenge which kicked off the race for self-driving cars. (His team placed third.)
Asked on Twitter whether his lab was working with Tesla, Hong posted some playful emojis but otherwise declined comment. We may learn more about how Hong’s work and Tesla’s pursuits intersect during AI Day.
Musk has been forthcoming about his desires for Tesla to become more than just a car company. “I think long term, people will think of Tesla as much as an AI robotics company as we are a car company or an energy company,” he said earlier this year.
A warning for anyone tuning in to the AI Day livestream: take Musk’s predictions about near-term accomplishments with a massive grain of salt. The things that will be discussed during this event are unlikely to have any measurable impact on the company’s business in the months to come.
Self-driving cars are an incredibly difficult challenge. Even companies like Waymo that are perceived to have the best autonomous vehicle technology are still struggling to get it right. Tesla is no different.
“A key question for investors will be what the latest timeline is for achieving full autonomy,” Loup Funds managing partner Gene Munster said in a note. “Despite Elon’s ambitious goal of the end of this year, our best guess is that 2025 will be the first year of public availability of level 4 autonomy.”
The rest of 2021 is already jam packed for Tesla. The company needs to open factories in Texas and Germany. And it needs to tool up production for its hotly anticipated Cybertruck, which has been delayed until 2022. Full autonomy, such as it is, can wait.