Elon Musk said that Tesla could have fully self-driving vehicles by May 2023. This is the latest in a line of ambitious timelines from Musk, who has sometimes been criticized for underestimated the amount of time it would take to accomplish a complex goal like creating autonomous vehicles.
If Musk gets the timeline wrong, it wouldn’t be the first time he had to scale back an ambitious goal. He previously aimed to have 1 million self-driving “Robotaxis” on the road by 2020. Then he said that he aims to have 1 million vehicles with the Full Self-Driving software in the hands of Tesla owners.
Tesla’s driver assist programs, Full Self-Driving (FSD) and Autopilot, have been the subject of scrutiny from regulators and authorities investigating wrecks involving a Tesla vehicle. Tesla maintains logs of FSD and Autopilot usage on its central storage that can be turned over to authorities if necessary.
The NHSTA opened a probe into the Autopilot’s apparent inability to recognize stopped emergency vehicles. Tesla had issued an update to make Tesla Vision capable of recognizing emergency lights but that didn’t come fast enough to prevent at least eleven reported accidents involving a Tesla vehicle and an emergency vehicle.
Autopilot logs have established that the driver assist software wasn’t at fault in accidents in the past despite initial police reports that indicated that Autopilot was a “suspect.” Users say that FSD has, on the whole, improved its ability to handle most driving tasks with each new update despite occasional setbacks like FSD Beta 9.2. Even Elon Musk admitted that Beta 9.2 was “actually not great” after taking it for a whirl.
The driver assist programs still include a disclaimer that drivers should remain alert and keep their hands on the wheel at all times. A driver can still be held legally liable for an accident that occurs while the driver assist program is active.
Tesla uses AI and more than a billion miles’ worth of driving data sent from its customers’ vehicles in real-time to “train” its driver assist programs. In June 2021, it spun up the world’s fifth most powerful supercomputer – which, despite some impressive specs, still isn’t as powerful as Tesla’s planned Dojo supercomputer.
“For us, computer vision is the bread and butter of what we do and what enables Autopilot. 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,” Tesla AI chief Andrej Karpathy told an audience at the 2021 Conference on Computer Vision and Pattern Recognition.
Tesla owners have expressed confidence in the abilities of Autopilot and FSD, though they may have occasionally gone a bit overboard in their confidence. One Californian Tesla vehicle owner was busted riding in the back seat of his vehicle with Autopilot engaged as a stunt. The stunt likely prompted Tesla to activate a camera mounted to the rearview mirror to track driver alertness.
However, Musk may still be overly ambitious in his goal of having fully autonomous vehicles by 2023. The problem is still, of course, humans and their often unpredictable driving habits, which can fake out a computer.