SoylentNews is people
SoylentNews
Lots of companies are working to develop self-driving cars. And almost all of them use lidar, a type of sensor that uses lasers to build a three-dimensional map of the world around the car. But Tesla CEO Elon Musk argues that these companies are making a big mistake. "They're all going to dump lidar," Elon Musk said at an April event showcasing Tesla's self-driving technology. "Anyone relying on lidar is doomed."
"Lidar is really a shortcut," added Tesla AI guru Andrej Karpathy. "It sidesteps the fundamental problems of visual recognition that is necessary for autonomy. It gives a false sense of progress, and is ultimately a crutch."
In recent weeks I asked a number of experts about these claims. And I encountered a lot of skepticism. "In a sense all of these sensors are crutches," argued Greg McGuire, a researcher at MCity, the University of Michigan's testing ground for autonomous vehicles. "That's what we build, as engineers, as a society—we build crutches."
Self-driving cars are going to need to be extremely safe and reliable to be accepted by society, McGuire said. And a key principle for high reliability is redundancy. Any single sensor will fail eventually. Using several different types of sensors makes it less likely that a single sensor's failure will lead to disaster.
"Once you get out into the real world, and get beyond ideal conditions, there's so much variability," argues industry analyst (and former automotive engineer) Sam Abuelsamid. "It's theoretically possible that you can do it with cameras alone, but to really have the confidence that the system is seeing what it thinks it's seeing, it's better to have other orthogonal sensing modes"—sensing modes like lidar.
Previously: Robo-Taxis and 'the Best Chip in the World'
Related: Affordable LIDAR Chips for Self-Driving Vehicles
Why Experts Believe Cheaper, Better Lidar is Right Around the Corner
Stanford Researchers Develop Non-Line-of-Sight LIDAR Imaging Procedure
Self Driving Cars May Get a New (non LiDAR) Way to See
Nikon Will Help Build Velodyne's Lidar Sensors for Future Self-Driving Cars
(Score: 5, Informative) by edIII on Wednesday August 07 2019, @11:10PM (24 children)
Lidar is not doomed, nor does it "sidestep the fundamental problems of visual recognition".
It's a SENSOR. Using several different types to provide input to the AI programs only makes sense. Cameras don't have depth perception, while lidar does. Using multiple cameras from multiple angles can help get you that data using image reconstruction algorithms. That's how a Kinect works. How is determining distance to an object superior by using AI reconstruction methods on static images than using an actual farking laser beam? Last I checked lasers could actually determine surface temperature of an object too.
These technologies complement one another in the same way a human body can detect heat and pressure. It's all just information, and were feeding it to AI anyways. More information, and more redundant systems for providing that information, can only aid the AI in its task. It's like a human being that may be temporarily blinded, but can still attempt to navigate by feel, touch, and sound.
Elon Musk is just being a dick trying to discredit other navigational systems while hawking his own crap as superior.
Technically, lunchtime is at any moment. It's just a wave function.
(Score: 3, Funny) by Anonymous Coward on Wednesday August 07 2019, @11:21PM (2 children)
Mod parent down. Elon Musk is not an arrogant prick, he is a gift from God to all mankind.
(Score: 0) by Anonymous Coward on Thursday August 08 2019, @01:03AM (1 child)
* [citation needed]
(Score: 1, Funny) by Anonymous Coward on Thursday August 08 2019, @01:19AM
https://twitter.com/elonmusk [twitter.com]
(Score: 2, Informative) by Runaway1956 on Thursday August 08 2019, @12:24AM (6 children)
Musk should take at least a brief look at warships. The typical warship has radar, of course, but looking more closely, there is repetitive redundancy obvious in the masts. Multiple straight antennae, along with multiple radar domes, each with intimidating name. Ships have several radio antenna on those masts as well. Sonar is redundant, in that multiple different frequencies are used, each in several different ways. As if that weren't enough, one of the ships I served on had a passive sonar fish that was towed behind the ship, on miles of cable, that could dive below the thermocline. Repetitive redundancy is everywhere on a warship.
Why all of that redundancy? It's to make the ship SURVIVABLE, both in peacetime and in combat.
Few motorists would consider survivability a "bad thing".
Stick those damned "extra" sensors in there. The car already costs multiple tens of thousands of dollars. Lidar increases the cost by $800 per vehicle? Don't be a cheapskate, put it in there!
(Score: 0) by Anonymous Coward on Thursday August 08 2019, @12:39AM (4 children)
Lidar adds approximately $10 to $100,000 to the cost of the vehicle.
(Score: 2) by Mykl on Thursday August 08 2019, @12:48AM (2 children)
Am I right in assuming that you meant $10,000 to $100,000?
(Score: 1, Interesting) by Anonymous Coward on Thursday August 08 2019, @12:59AM
NO! [spar3d.com]
(Score: 2) by deimtee on Thursday August 08 2019, @01:03AM
Probably not.
A single distance sensing laser (eg cheap handheld distance meter) can probably be sourced for less than $10 per unit.
A full-on 'warship' LIDAR array as described by Runaway above could easily go above $100,000
If you cough while drinking cheap red wine it really cleans out your sinuses.
(Score: 2) by Runaway1956 on Thursday August 08 2019, @01:36PM
Y U HATE CAPITALISM????
Just because you can find a chip for ten bucks, doesn't mean you will find anyone who will install that chip for you, along with all the rest of the hardware and electronics to make the chip useful for ten bucks. I picked $800 as a semi-rational figure, because I know it's not super expensive, but no one is going to do it for free.
(Score: 2) by c0lo on Thursday August 08 2019, @02:54AM
As a motorist, I confirm I'd like to see cars with miles of cable in tow if those cars are more survivable ⚆_⭕ (grin)
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 2) by legont on Thursday August 08 2019, @02:38AM (7 children)
Not trying to get into lidar vs whatever discussion, more information does not mean better decision. It is good to have say vision and hearing, while it is not good to have two visions.
"Wealth is the relentless enemy of understanding" - John Kenneth Galbraith.
(Score: 0) by Anonymous Coward on Thursday August 08 2019, @02:55AM (1 child)
(Score: 1) by khallow on Thursday August 08 2019, @12:32PM
Yet.
(Score: 3, Touché) by c0lo on Thursday August 08 2019, @02:56AM
Ummm... binocular vision, not good?
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 0) by Anonymous Coward on Thursday August 08 2019, @10:44AM
Have you had one of your eyes removed yet?
(Score: 2) by Runaway1956 on Thursday August 08 2019, @01:37PM
So - you would prefer to have bat sonar, rather than lidar? QUICK - to the Batmobile, and I'll show you how it works!
(Score: 2) by Runaway1956 on Thursday August 08 2019, @01:43PM (1 child)
Let's say you've got six sensing systems. You can figure that one goes out now and then. So, fail-safe it. If three or more of your systems agree that you're good to go, you can go - at reduced speed. Nobody gets stranded at the side of the road because one out of six isn't happy.
Back aboard ship again - we had four big-ass boilers for main power. If one boiler was sick, we ran with three. That's how repetitive redundancy is supposed to work.
Your autonomous vehicle should have lots of different sensors, and the car should only shut down if multiple sensors agree that conditions are actually unsafe.
(Score: 2) by legont on Thursday August 08 2019, @07:06PM
The only thing I wanted to point out is that a statement "two systems are better than one" is often not true. When two systems provide similar information the decision is often worse than with one system. It does not mean that a backup is bad. It does not even mean that having two system is necessarily bad. It simply means that a designer should be ware of the fact that adding a sensor may decrease the results, which the original poster is not aware of.
More generally, the philosophy "more is better" is typical for an American design and it is a weakness, I believe.
"Wealth is the relentless enemy of understanding" - John Kenneth Galbraith.
(Score: 0) by Anonymous Coward on Thursday August 08 2019, @02:47AM (1 child)
(Score: 1) by khallow on Thursday August 08 2019, @09:49AM
Yet. An obvious way to extend the dynamic range of a camera is via an iris, just like what the human eye has. And human night vision is based on chemical processes that aren't that reliable. You can ruin night vision for minutes to hours (depending on how light sensitive you're aiming for) by brief exposures to bright light (such as headlights from a passing car).
(Score: 1) by khallow on Thursday August 08 2019, @09:43AM (2 children)
To be fair, it's a combination energy projection and sensor system, an active system as opposed to a passive system which doesn't require such energy projection (except under the usual low light conditions that a human driver would experience as well).
As to more sensors complementing each other, let us not forget the large, synergistic downside - more failure modes. All this redundancy sounds great, but what happens when it creates more simple opportunities for a car to become undrivable? I think that ultimately is the catch. The more "redundant" systems you add to a car, the more "This car can't drive" failure modes you create. And keep in mind that people are notorious for driving cars with serious problems. If they can drive with broken sensors (and thus, reduced redundancy), they will.
Presently, supplementary external sensors to a human driver aren't essential. They can to the very last one be disabled without creating a car that can't drive. But with an autonomous car, those additional sensors create additional liability. In particular, whatever else you can say about lidar, it remains that it has two components that need to work in order for it to function rather than the one component of a passive video system.
A vehicle with a small number of critical sensors (that is, where a single failure makes the car undrivable for either safety or liability reasons) is going to have better operational reliability than a system with lots of critical sensors, even if all those sensors are individually somewhat more reliable. If I double the number of critical sensors (assuming all have equal rate of failure), I need to halve the likelihood of failure of those sensors in order to maintain the same reliability.
I think that's the calculus behind Tesla's decision.
(Score: 0) by Anonymous Coward on Thursday August 08 2019, @12:31PM (1 child)
fail on ignorance--
Tesla currently uses both cameras and radar (passive and active in your usage). Here's a Tesla forum post on the radar, https://teslamotorsclub.com/tmc/threads/where-is-the-front-radar-on-hw2-cars-located.87415/ [teslamotorsclub.com]
If you look down a few posts, it appears that the cruise control in some models of Tesla depends on the radar working...and the radar doesn't work if covered with snow/ice.
(Score: 1) by khallow on Friday August 09 2019, @12:41AM
Still means two more systems that have to work on top of what they already have. And as I noted, now you need five systems to work, not three.
In other words, a failure mode. I mentioned that happens. It's not like a lidar equipped car would be racing along in those conditions either.
(Score: 2) by digitalaudiorock on Thursday August 08 2019, @02:37PM
Correct me if I'm wrong, but it sure seemed like that first widely reported self-driving fatality appeared to be because his beloved "visual recognition" couldn't possibly see a white semi-truck against a white background? Sounds more like Elon is trying to "sidestep" the cost of a real sensor(??).