A shortsighted design decision that Elon Musk made more than a decade ago is once again coming back to haunt Tesla.
As The Guardian reports, the company's Robotaxi rollout has been a massive bust — due, in at least in part, to Musk's long-ago bet against the light detection and ranging sensors known as lidar, which are hardware that allow cars to "sense" their surroundings far more sensitively than the visual cameras that Tesla is instead using as the inputs for itse self-driving software.
Waymo, Google's driverless cab service, uses a mixed array of lidar, radar, cameras, and other sensors to autonomously navigate the world around them.
More than decade ago, Musk made a big bet against that strategy. And he's kept raging against it since, too.
During an October 2015 press conference about Tesla's Autopilot assisted driving feature, the CEO called lidar "unnecessary" and said the tech"doesn’t make sense" for an autonomous car. (Notably, he also claimed in that same presser that full autonomy would "obviously need 360 cameras," though he's seemingly changed his tune since.)
"Lidar is lame," the CEO insisted at a Tesla autonomy day event back in 2019. "In cars, it's friggin' stupid. It's expensive and unnecessary."
He's right about the price tag. As the Guardian notes, a suite of lidar sensors runs about $12,000 per vehicle, compared to the $400 it costs to install cabin cameras.
Still, safety should matter more than cost. Given that Tesla's Robotaxis have already been seen braking repeatedly around cop cars, jerking a steering wheel back-and-froth, and breaking so many traffic laws on camera that the government has taken interest in the single week since the service's launch, Musk's line in the sand seems more foolhardy than ever.
In certain cases, it's even been directly implicated in deadly accidents. In a deadly 2019 crash, a Tesla on Autopilot failed to differentiate the side of a large truck from the scenery behind it. And in a fatal 2023 crash, a Tesla Model Y seemingly got blinded by the Sun and ran down a grandmother standing in the roadway.
The issue is that lidar essentially "scans" the environment around a car, giving its self-driving system access to an extremely accurate internal map of its surroundings. Visual cameras show what a human driver would see, but the flat visual data is much more complex to interpret and far more susceptible to hard-to-predict edge cases involving lighting conditions and other confusing factors.
... continue reading