Tesla Autonomy Investor Day: What we learned, what we can look forward to
Elon Musk revealed Tesla's Full Self-Driving tech to shareholders, along with a plan for a test network of 1 million unmanned robotaxis by the end of 2020.
Chris PaukertFormer executive editor / Cars
Following stints in TV news production and as a record company publicist, Chris spent most of his career in automotive publishing. Mentored by Automobile Magazine founder David E. Davis Jr., Paukert succeeded Davis as editor-in-chief of Winding Road, a pioneering e-mag, before serving as Autoblog's executive editor from 2008 to 2015.
Chris is a Webby and Telly award-winning video producer and has served on the jury of the North American Car and Truck of the Year awards. He joined the CNET team in 2015, bringing a small cache of odd, underappreciated cars with him.
Tesla CEO Elon Musk revealed a number of details regarding the company's new Full Self Driving tech at today's Autonomy Investor Day seminar. The Silicon Valley company's plan to detail its driverless tech comes amid growing questions about global demand for
electric vehicles, the ongoing health of its stock, and the stability of the company's leadership -- Musk included. In other words, this promised to be a hugely important event in the history of the high-profile automaker -- and based on some of the claims that Musk made, that looks to be the case.
Back on April 3, Tesla first announced plans to host Autonomy Investor Day, which it originally scheduled for Friday, April 19. It subsequently moved the date to Monday, April 22, "based on investor feedback."
According to Tesla's initial press announcement, the company "is making significant progress in the development of its autonomous driving software and hardware, including our FSD computer, which is currently in production and which will enable Full Self-Driving via future over-the-air software updates." On April 15, Musk upped the ante by tweeting, "Buying a car in 2019 that can't upgrade to Full Self-Driving is like buying a horse instead of a car in 1919."
Tesla also took attending investors on test drives to experience its latest iteration of Autopilot, including demonstrating features and functions that remain the subject of ongoing development.
Tesla -- and
in particular -- have been under fire for promising full autonomous-drive technology at a time when other automakers and suppliers seem to be backing away from ambitious timetables, lengthening their production horizons to account for growing technological challenges.
For its part, Tesla's self-named Full Self-Driving hardware doesn't seem to include core technologies that virtually every other company racing towards driverless cars has deemed necessary for safe, reliable operation. Musk stated during the presentation that he believes Tesla's Full Self-Driving system will be feature complete by the end of 2019. He then went on to state that he anticipates that the system will be significantly robust to permit users to not pay attention while using it by Q2 of 2020. Finally, Musk said that he believes Tesla will be in-process on receiving regulatory approval from the US government by the end of 2020.
Tesla's Full Self-Driving suite will rely on a network of cameras, as well as forward-facing radar and ultrasonic sensors to keep tabs on a vehicle's immediate surroundings. Every other major player in this growing space, including automakers like Audi and tech companies like Mobileye and Waymo are developing prototypes that employ lidar sensors (laser radar), as well as hyper-detailed three-dimensional mapping. Tesla has not shown such technologies, and Musk has been dismissive of lidar technology in the past. Instead, Tesla has unveiled a major step forward in the form of a new AI chipset developed in-house to power its FSD tech.
Tesla pulls the wraps off its Model Y crossover SUV
Leaving out the sensors that Tesla plans to use to facilitate self-driving, its hardware solution seems incredibly robust. The Full Self-Driving (FSD) computer package utilizes redundant, custom-designed chips to deal with the massive amounts of information provided by the rest of the hardware package. It is designed from the ground up by Tesla to do this while using a minimal amount of power (under 100 watts in this case) to reduce the system's impact on overall vehicle range.
The current generation of FSD hardware has been finished for approximately two years, and the next generation of hardware -- according to Musk -- is due out in another two years. Musk predicts that the next iteration of FSD hardware will be approximately three times as powerful as the current one. Currently, all new Teslas being built come standard with this FSD computer hardware installed, and many older vehicles can have it retrofitted with minimal effort.
Neural network and software
Tesla's Navigate On Autopilot hardware that's currently available on its vehicles is already capable of fully automatic lane changes and finding its way onto and off of freeway interchanges. The system has not been without its critics, and there have been a number of high-profile accidents that some have attributed to blind spots in the technology's makeup. Despite those incidents, there's no denying that Autopilot offers more functionality in more environments than competing systems currently available from other automakers (many of which have been more conservative in deploying new features for safety reasons).
One of the main driving factors that Tesla is relying on to move its self-driving tech forward is its neural network. The Tesla neural network pulls images and video from Tesla vehicles in the real world and through a combination of human annotation and predictive behavior, it learns how to read the road efficiently and accurately. It also uses the vehicle's radar sensors to help it determine object depth and distance without needing to use a stereoscopic camera setup. Tesla claims that it has upward of 425,000 cars on the road with Hardware 2.0 that are uploading data for training to be used by the neural network.
Andrej Karpathy, Tesla's head of AI also expounded on Musk's statements that lidar is a fool's errand and a crutch. He explained that the Tesla system uses computer vision because the world's roads are built on vision, and using a neural network to read images offers dramatically more information for a given object than can lidar.
Elon Musk portended during the presentation that, "Anyone relying on lidar is doomed."
While the Autonomy Investor Day event was centered on self-driving tech, we also got plenty of other nuggets from Musk and other Tesla officials regarding future products like the Model Y and the Semi. One particularly interesting tidbit was Musk's mention of a planned battery pack redesign that would allow it to last for around a million miles (current packs are expected to go between 300,000 and 500,000 miles before needing replacement).
We also got an earful about Tesla's planned robotaxi network. Aside from the technological requirements, we learned that Tesla already has designs on its off-lease vehicles. If you lease a Model 3, you won't have the chance to buy it at the end of your contract; Tesla wants it back. This will help the company build up its standing fleet for a future robotaxi service in areas that don't have enough owners wiling to share their vehicles. Musk plans on having over a million robotaxis operating on the road as of 2020, though with nobody in them.
Whether you're a Tesla bull or bear, believer or cynic, Autonomy Investor Day was packed with interesting information, and we look forward to more presentations like this from the brand in the future.
Originally published April 22, 6:39 a.m. PT. Update, 2:18 p.m. PT: Added further information regarding Tesla Full Self-Driving.
Correction, 3:36 p.m. PT: A previous version of this story said that Elon Musk planned to have over 1 million robotaxis operating by the end of next year. That was an incomplete quote, and has been adjusted to reflect Musk's actual statements.