Do self-driving cars need LiDAR? Elon Musk says no, but most other experts say yes. And we’re going to talk to one of them today
In a recent episode of TechFirst with John Koetsier, Omer Keilaf, founder and CEO, Innoviz, who supplies lidar for BMW and other manufacturers, says LiDAR is essential, because water or mud or dust can disrupt visual sensors.
And, of course, all this is happening in an era when LiDAR is getting so cheap we can have it in our phones. Right now it’s at $1000 for automotive uses; that’s coming down to $100 or even less.
Scroll down to subscribe to the podcast, get the full video, and read the transcript …
And … here’s the story on Forbes …
TechFirst: is Elon Musk wrong about LiDAR?
Watch: do we need LiDAR in self-driving cars?
(Subscribe to my YouTube channel so you’ll get notified when I go live with future guests, or see the videos later.)
Read: LIDAR, self-driving cars, and BMW supplier Innoviz
(This transcript has been lightly edited for length and clarity.)
John Koetsier: Do self-driving cars need LIDAR? Elon Musk says, no way. Most other experts say, probably. We’re going to talk to one of them today. His name is Omer Keilaf, and he’s the founder and CEO of Innoviz, who supplies LIDAR and other technology for BMW and many other manufacturers.
In an era, of course, when LIDAR is getting so cheap, we can have it in our phones. Welcome, Omer!
Omer Keilaf: Thank you. Thank you for inviting me.
John Koetsier: Good to have you here. I’m looking forward to this conversation. Let’s start right here: Do we need LIDAR to make self-driving cars safe? If so, why?
Omer Keilaf: Yeah, absolutely. I mean, an autonomous car is driven by a computer and you can’t allow it to have a single point of failure, right?
So, you know, a drop of water can blur a camera and then the computer can not understand the scene. Today, you have a passenger which is holding a wheel and taking over if something wrong happens. But if you want to really fully disengage, you need to add another sensor that is not degraded in the same way that cameras are, that can actually provide redundancy. Today, the only sensor that can provide it is a LIDAR. So if you want to get to autonomous driving, you have to use a LIDAR.
John Koetsier: Big elephant in the room … Elon Musk and Tesla have been … hardcore, let’s put it that way, that LIDAR is unnecessary. It’s almost a religious position it seems in some way, shape, or form. Why?
Omer Keilaf: Yeah. I mean, five years ago, when he had to make a decision on a car that he brings to the market, there was no LIDAR that was available at the right price point and performance and maturity.
So he made a good decision.
I think the reality has changed. I think that the people realize that LIDARs are able to reach the relevant price point and performance. And basically, I believe eventually he would make the right decision if he wants to go for Level 3.
John Koetsier: So we’ll get into some of the costs and cost curves and where it’s going, where it’s trended in a little bit. Talk to us a little bit about how much better LIDAR is. You mentioned already that it doesn’t get blinded by a drop of water or some mud on the sensor or something like that that might come up from the road.
How much better is LIDAR — what’s the resolution, the range, those sorts of things — than visual?
Omer Keilaf: Yeah. So, range and resolution are calculated by the speed of the car that you want to drive autonomously. The faster the car—
John Koetsier: Very fast. Very, very fast. [laughing]
Omer Keilaf: Yeah. So, for example, when a car drives at 130 kilometers an hour, it has a certain time it needs to slow down, right? So you can actually calculate the range which you need to see. So about 200 meters is kind of the figure of merit in that. And you need to have very high resolution, because you want to be able to see an object which is at least one-third of the tire, because that’s an object that can actually make a car turn over. So the resolution is quite high, it’s 0.1 or even higher than that.
And you need to have first a high frame rate in order to react faster and wide field of view in order to capture different driving scenarios such as incline or decline or turns, etc. And, you know, cameras can achieve quite well in seeing objects, as long as they see. And when they don’t, there is no other sensor that can reach that. Even the most advanced radars that are in discussion today are far below the needed resolution. So, it’s only LIDARs.
John Koetsier: Okay. Now, you’re working with BMW, you’re installing them, your LIDARs, in their cars. What are they using it for? And how’s that project going?
Omer Keilaf: Yeah, sure. So this is our product. As you can see, it’s very, very small. This is InnovizOne, it’s a solid state LIDAR. It’s mounted in the car grill in order to provide that redundancy information, also including object detection and classification. So we’re not only providing the raw 3D information, we also provide insights — meaning that we translate the road data into cars, pedestrians, motorcycles — and that gives a redundancy to the overall system.
The project is quite interesting. Being part of such a program is a big honor for us. Now BMW is one of the most, I would say, advanced car makers in regards to the technology … and we’re very proud to be picked for that.
John Koetsier: So let’s talk about those cost curves that you mentioned earlier. I remember when LIDAR sensors were thousands of dollars — I even want to say tens of thousands of dollars — and now, of course, you’re getting them in smartphones; you’re getting them in smaller devices.
Talk to me about the evolution there and where the cost curves have come, and where they’re going.
Omer Keilaf: Yeah. I mean, traditionally whenever a new technology is introduced to the market, those are mostly adopted by the premium cars, because they have sufficient volume but can absorb a certain cost.
So today, for a premium car, a LIDAR is about $1,000, and I think it still meets the market demand. But if you want to go into a lower grade, the LIDAR needs to be much cheaper, right? So we all know that at the really high adoption of LIDARs, it will need to be at $100 or even less.
I think that in the next five or seven years, probably the trend would go through $500 and then $300. And this is where we are targeting our next generation: InnovizTwo. I’ll give you an interesting insight. You know, five years ago, there was a big discussion whether you use stereo cameras or mono cameras, right?
And Mobileye has proven [to] the world that even saving tens of dollars — right, the difference between two cameras and one, is only $10 or $20 — can be compensated by a very strong IP. I think that the push for lower cost LIDARs would be very, very harsh. I think that using our kind of technology is the right way to get there.
John Koetsier: So what are you using? Are you using multiple cameras or one?
Omer Keilaf: We’re using one. [crosstalk & laughing] And in that benchmark, we’re using one. I mean, there is a dispute in the LIDAR space between using 1550 nanometer and 905 nanometer. We are on the 905 camp, because that’s allowing us to use very low cost components, and we’re able to solve performance limitations of 905 by very strong IP that we developed.
John Koetsier: Okay. So we’re talking self-driving cars, obviously. You’re talking BMW, one of your customers. Those are one application of using LIDAR; one application of autonomy, frankly, for machines or even cars, other types of transportation. How else, and where else do you see it being used?
Omer Keilaf: Hmm, many — so, you know, there are different application verticals you would say, for LIDARs. One of them is shuttles. So those are urban creatures that like to travel in a predetermined route.
And there is a lack of drivers also in trucks, so they need — the use of LIDARs can actually quite be beneficial in order to improve transportation as a whole. So trucks, shuttles … of course also there are other industrial applications that would like to automate. I mean, anything you’d like to automate would like to have 3D sensing, because it gives you a very good understanding of the scene, and would be able to be enabled by a LIDAR.
I think a LIDAR has a very strong benefit which gives you a very good understanding of the scene for very different types of objects. A camera is mostly relying on the fact that you are able to train the system to identify what it is in order to interact with it. A LIDAR is actually a physical sensor. You send a pulse of light and you measure a reflection, a light reflection from an object. So you basically can see it. So you know where it is. You know what’s the size of it, even if you don’t exactly know what it is. I will talk about the example of a person pushing a piano into the road … that might happen, right? I mean, if you didn’t train your camera to understand what is that object now pushed to the road, not sure how would you react to it.
And with a LIDAR, you basically see where it is; you see how large it is; how far you are with it; and you know how to walk around it. So it’s not only about making it safer, it’s actually also providing lots of insights that are missing from the camera, and the processing power that you need to add in order to understand the scene is going to be much lower.
John Koetsier: So Apple is rumored to be entering the self-driving car market, perhaps 2024, and they’re talking about using LIDAR as well, from what we hear. How do you see rumors of that impacting your industry — the car industry — and if that turns out to be true, what do you think the impact will be?
Omer Keilaf: I think it’s quite good, obviously. It’s quite clear why Apple has an interest to go in that direction. Tesla has showed the world that being a very technology company allows them to be the most valuated company in the world. I mean, and it’s not because they’re selling more cars than others. I think it’s because of the fact that people believe that eventually they will gain more and more market share because of their technology lead.
Apple is a very strong technology player, and most likely would be a very strong player in the market if they decide to participate. And I think it’s a smart move. I think it helps also for us, as a supplier in the automotive space that pushes technology, to get more interest from car companies that want to be in the game, want to be more technology-oriented, and get faster to technology.
John Koetsier: It’s going to be so interesting. Obviously there’s talk about partnering. There was talk about some Japanese partners as well, that seemed to fall through. And I’m not going to ask you to speculate on that. Obviously you’ve got partnerships and interests in the industry as well, but Apple has a big war chest and it can make a lot of things happen whether it goes on its own, whether it partners, whether it actually purchases a full-on existing automotive manufacturer right now as well.
In terms of LIDAR, and in terms of devices that a vehicle has to sense its environment, how many should a vehicle have?
Omer Keilaf: So it depends on the application. For Level 2+ or Level 3, which is an autonomous car but mostly on highway, or traffic jams, it’s basically one sensor: one front-looking sensor. Obviously with sufficient field of view, okay, but it’s front-looking.
Once you go into an urban scenario, like a shuttle or a robo taxi, you need to have 360°, meaning that you need several LIDARS around the car. I think that in shuttles you might need even more, because — and I think that using a LIDAR such as ours has a quite interesting value, because shuttles are usually very long vehicles and very tall, so if you add a spinner to the roof, it will most likely meet like the roof from the top and will have a very huge blind spot around the vehicle.
John Koetsier: Yes.
Omer Keilaf: So that’s not very helpful. So anyway you need multiple sensors, even if they’re 360°. And that’s not very efficient and it’s quite expensive. So having a LIDAR like ours, which is very small and you can embed the surrounding on the vehicle and you can place them lower, gives you the benefit of using a lower cost automotive grade and short range, and long range. So, I think that that’s a very interesting market for us.
John Koetsier: Yeah. Yeah. Very interesting. So TechFirst is about tech that’s changing the world and innovators who are shaping the future. Where do you think the future of machines and perception is? How do you see that evolving over the next 5-10 years?
Omer Keilaf: I believe it’s going to be growing very fast. I think that once the — we all see that there are two trends that are supporting that kind of application. One is the processing power, and the second is the sensors. And, you know, I’m also amazed by the pace that we are able to achieve. You know, with InnovizTwo, which will be released later in the year, it’s going to be a product that is going to be mind blowing. And it’s quite amazing to see how technology can run so fast.
And, to your point earlier about Elon Musk, which is very religious about LIDARs, you know, I always say that I don’t really understand why people would be religious on technology. They tend to change so fast. I think that they change faster than we can realize. And obviously while we’re talking about also quantum computing, it’s really going to be quite amazing. You know, it’s funny — I don’t know if you know, but in automotive there is a lot of work on cybersecurity.
John Koetsier: Mm-hmm.
Omer Keilaf: And one of the things that we need to meet with is with the chance that in the next 15 years, quantum computers would be available to attack, you know, cyber security. So you need to prove that your cyber security today is going to be strong enough for the next 15 years. Think about that as a challenge.
John Koetsier: It’s a real challenge. And of course you can update over the air, but the hardware is not so easy to update. And we see that coming very soon, right?
I mean, Teslas are already in an unreleased version talking to each other, telling each other what’s up ahead on the road, prepping to go and caravan together, right? So they break the wind and you are 20% more efficient — battery efficient, I should say, rather than fuel efficient — traveling.
So it’s super interesting. So you’re releasing version two later this year. Where do you see your technology in … let’s say a decade? Is it too small to see? Is it tiny? Is it invisible in the car? Perhaps ubiquitous? Ten little pods around the car that you don’t even notice?
Omer Keilaf: Yeah, it’s a good question. I think that, you know, it’s clear that there are two ways to solve it. One is that you need to have a sensor which is cheap enough that you can embed in any car and actually multiple of those. And then there’s the infrastructure. So I don’t know which one of them would come faster. I assume that the earlier one, because even when the infrastructure is going to be available, it will have its own limitations.
I believe that LIDARs would be very much smaller and much cheaper, like sub $100 and even lower than that, and I have no doubt it will get there.
John Koetsier: Excellent. Thank you so much for your time.
Omer Keilaf: Yeah, sure. It was a pleasure. Thank you for inviting me.
Read all that? Just subscribe already …
Made it all the way down here? Who are you?!? 🙂
The TechFirst with John Koetsier podcast is about tech that is changing the world, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer (yeah, that’s this one!). Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.
Subscribe on your podcast platform of choice: