The autonomous driving world is still buzzing about Tesla’s “Autonomy Day” last month, where CEO Elon Musk and a host of others tried to convince everyone that Tesla’s self-driving cars will be “feature complete” before the end of 2019 and that its fleet of robot taxis will be ready for regulator approval and full roll out in certain markets “sometime next year” — all with the use of radar and imaging, not lidar.
Elon: “There’s 3 steps to self-driving: there’s feature complete, then there’s feature complete to the degree that … where we think that the person in the car does not need to pay attention, then there’s at a reliability level where we also convince regulators that that is true.
“We expect to be feature complete in self driving this year, and we expect to be confident enough from our standpoint to say that we think people do not need to touch the wheel and can look out the window sometime probably around … in the second quarter of next year. And we expect to get regulatory approval, at least in some jurisdictions, for that towards the end of next year. That’s roughly the timeline that I expect things to go on. …
Of course, Musk’s disdain for lidar is nothing new. We told you three years ago that Musk was doing his 3D sensing with cameras and radar (even making point clouds with radar), but now he’s doubling down and beginning to win over converts. The question is: Does Musk’s position matter for the lidar market at large?
Well, if the most prominent naysayers are correct and Tesla has bet incorrectly, it could be a major boon to lidar manufacturers and the mainstreaming of the technology in general. Just as Olivia Solon argued in the Guardian in 2016, lidar’s biggest defenders today continue to point to the irresponsible nature of going lidar-less, saying only lidar can deliver the safety necessary to make driverless cars able to navigate roadways without causing dangerous accidents.
Here’s AI writer Lance Eliot in Forbes this week:
When an autonomous car gets into an accident, which mark my words will happen, you can be further assured that lawsuits will be filed. I’ve previously described that one of the key elements of such lawsuits will be what did the autonomous car maker do as part of the design, building, and fielding of their self-driving car, and particularly how did they seek to ensure safety and reliability.
For Tesla, if pressed by a lawsuit, they will need to defend in court their decision to not use LIDAR. As you can imagine, Tesla will be on rather shaky ground if it is shown that essentially all other autonomous cars are using LIDAR. The burden to explain and justify the lack of LIDAR on Tesla’s is going to be mighty steep.
The Tesla position would presumably be that the addition of LIDAR would not have materially avoided the car accident and loss of lives, but this is going to be tough to showcase since in theory any use of LIDAR is going to incrementally improve the safety odds, assuming it is used wisely, and so it’s another part of the uphill climb by Tesla to avoid getting summarily dinged for their lack of LIDAR.
So, Tesla gets smoked in a lawsuit due to lack of lidar and all of a sudden the company has to retrofit its plans with lidar data? Boom. That’s a lot of new lidar units needed.
But this begs the question, doesn’t it? Both the Guardian and Eliot posit that “in theory any use of lidar is going to incrementally improve the safety odds.” But is that actually true? We don’t know! Isn’t it possible the use of cameras and radar is just as effective? People have been arguing for photogrammetry as a cost-efficient alternative to lidar since I’ve been in this market. Isn’t it possible Tesla has figured out how to do the AI faster and better with images and radar?
The claim to have figured it out, anyway, with pointed commentary:
“Lidar is a fool’s errand,” Elon Musk said. “Anyone relying on lidar is doomed. Doomed! [They are] expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices. Like, one appendix is bad, well now you have a whole bunch of them, it’s ridiculous, you’ll see.”
Oh, and Andrej Karparthy, Teslas senior director of AI, piled on:
“In that sense, lidar is really a shortcut,” Karparthy said. “It sidesteps the fundamental problems, the important problem of visual recognition, that is necessary for autonomy. It gives a false sense of progress, and is ultimately a crutch. It does give, like, really fast demos!”
What’s unclear is why not just use lidar in combination with the imaging instead of radar. Tesla seems to be clinging to the idea that lidar is expensive, but at $250 a unit and a host of companies trying to push that lower, it’s hardly a show-stopper for a $37,000 car. It doesn’t seem possible that Tesla has just completely missed the massive decrease in price per unit, right?
Well, Musk has an answer for that. Sort of:
“If you’re gonna use active photon generation, don’t use visible wavelength, because with passive optical, you’ve taken care of all visible wavelength stuff. You wanna use a wavelength that’s occlusion-penetrating, like radar. LIDAR is just active photon generation in the visible spectrum.”
That seems to be an apples-and-oranges kind of thing, but maybe you can parse it. Velodyne certainly isn’t impressed with Tesla’s claims. And they think Musk is muddying the overall waters, confusing people with what autonomy really is.
“Velodyne takes the position that ‘autonomy,’ implies that a car can drive safely and do all functions while a driver takes a nap,” [Velodyne President Marta Hall] said. “Tesla offers what are really ‘driver features,’ not autonomy. They are cool features that are fun, but not for napping.”
Further, you might be saying, who cares? There are any number of other automotive companies still all-in on lidar and one company going rogue isn’t exactly going to devastate the marketplace. Unless they win. Which at least one commentator is predicting (on Joe Rogan’s podcast, so take that with a grain of salt):
“If you’re going to build artificial intelligence systems…camera is the way to go, because you can learn so much more, you can see so much more,” Fridman said. “The richer, deeper sensor is the camera, but it’s much harder. You have to collect a huge amount of data. It’s a little bit more futuristic, so it’s a longer-term solution.”
As with the efficacy and applicability of lidar, Musk’s dependence on imagery is largely fueled by increased processing power. As part of their demonstration last month, Tesla touted a processor capable of churning through 2,300 frames per second: