Waymo's Fifth Generation Lidars, Radars and Cameras

[Bladerskb]

Waypoint - The official Waymo blog: Introducing the 5th-generation Waymo Driver: Informed by experience, designed for scale, engineered to tackle more environments

"With each generation of our custom hardware we’ve been able to bring down the cost of our sensors while delivering even more capabilities and compute power. With our fifth-generation hardware, we’ve simplified the design and manufacturing process so it can be production-ready, and our latest sensors deliver more performance than ever before, at half the cost of our previous generation. While lidar helps us see objects and cameras help us understand our surroundings, radar complements both of these with its unique ability to instantaneously see and measure an object's velocity (or lack thereof) even in tough weather conditions such as rain, fog, and snow."

LIDAR

As one of the Waymo Driver's most powerful sensors, lidar paints a 3D picture of its surroundings, allowing us to measure the size and distance of objects around our vehicle, whether they're up close or over 300 meters away. Lidar data can be used to identify objects driving into the sun on the brightest days as well as on moonless nights. Lidar lead Simon Verghese declined to divulge specs, but says the resolution on the new unit is 10 times better than its predecessor, thanks to “new architecture.”​

​

CAMERA

With high-dynamic range and thermal stability over automotive temperature ranges, our vision system cameras are designed to capture more detail and provide sharper images in the toughest driving environments. Our latest long range cameras and 360 vision system now see much further than before, allowing us to identify important details like pedestrians and stop signs greater than 500 meters away. Through advanced design innovations, including custom lenses and precise optomechanical engineering, our vision systems enable much higher performance levels than cameras on cars today.

​

RADAR

For our fifth-generation hardware sensor suite, we have redesigned the architecture, outputs, and signal processing capabilities to create one of the world's first imaging radar system for self-driving - providing us with unparalleled resolution, range, and field of view to see the whole scene at once. While traditional automotive radars are capable of tracking moving objects, our new imaging radar has higher resolution and enhanced signal processing capabilities that allow it to better detect and track objects that are moving, barely moving, or stopped.

Our next-generation radar can also see objects at great distances, including detecting a motorcyclist from hundreds of meters away. Like with our other long range sensors, being able to accurately detect objects at greater distances gives us a longer reaction time to make a more comfortable experience for our riders.​

They didn't show the output of their "imaging radar", but here's a illustrations in the industry comparing imaging radar to current automotive radar.

FULL ARTICLE
Waypoint - The official Waymo blog: Introducing the 5th-generation Waymo Driver: Informed by experience, designed for scale, engineered to tackle more environments

 

[diplomat33]

Waypoint - The official Waymo blog: Introducing the 5th-generation Waymo Driver: Informed by experience, designed for scale, engineered to tackle more environments

"With each generation of our custom hardware we’ve been able to bring down the cost of our sensors while delivering even more capabilities and compute power. With our fifth-generation hardware, we’ve simplified the design and manufacturing process so it can be production-ready, and our latest sensors deliver more performance than ever before, at half the cost of our previous generation. While lidar helps us see objects and cameras help us understand our surroundings, radar complements both of these with its unique ability to instantaneously see and measure an object's velocity (or lack thereof) even in tough weather conditions such as rain, fog, and snow."

LIDAR

As one of the Waymo Driver's most powerful sensors, lidar paints a 3D picture of its surroundings, allowing us to measure the size and distance of objects around our vehicle, whether they're up close or over 300 meters away. Lidar data can be used to identify objects driving into the sun on the brightest days as well as on moonless nights. Lidar lead Simon Verghese declined to divulge specs, but says the resolution on the new unit is 10 times better than its predecessor, thanks to “new architecture.”​

​

CAMERA

With high-dynamic range and thermal stability over automotive temperature ranges, our vision system cameras are designed to capture more detail and provide sharper images in the toughest driving environments. Our latest long range cameras and 360 vision system now see much further than before, allowing us to identify important details like pedestrians and stop signs greater than 500 meters away. Through advanced design innovations, including custom lenses and precise optomechanical engineering, our vision systems enable much higher performance levels than cameras on cars today.

​

RADAR

For our fifth-generation hardware sensor suite, we have redesigned the architecture, outputs, and signal processing capabilities to create one of the world's first imaging radar system for self-driving - providing us with unparalleled resolution, range, and field of view to see the whole scene at once. While traditional automotive radars are capable of tracking moving objects, our new imaging radar has higher resolution and enhanced signal processing capabilities that allow it to better detect and track objects that are moving, barely moving, or stopped.

Our next-generation radar can also see objects at great distances, including detecting a motorcyclist from hundreds of meters away. Like with our other long range sensors, being able to accurately detect objects at greater distances gives us a longer reaction time to make a more comfortable experience for our riders.​

They didn't show the output of their "imaging radar", but here's a illustrations in the industry comparing imaging radar to traditional automotive radar.

FULL ARTICLE
Waypoint - The official Waymo blog: Introducing the 5th-generation Waymo Driver: Informed by experience, designed for scale, engineered to tackle more environments

Thanks. Now that is what I call "FSD capable" hardware! It really puts Tesla to shame.

I love Waymo. They have excellent hardware and software and have shown to have real autonomous driving. They did 1.4 million real autonomous miles in CA last year with an excellent disengagement rate of 13,219 miles per disengagement.

 

[kevindamen]

Really not as good as it sounds.

Your looking at a camera system that is far more expensive then the car it goes on top of.

There is no point making a system that no one can afford to own.


Humans drive with two eyes and two ears.
Humans drive with a visual system of 20/60 or better. That is very crappy vision, and humans make it work.

Computers should be able to do the same with the same equivalents. (Maybe a few extra if the computer cannot turn to do shoulder checks, but only a few extra)

Anything more then that is just increasing the price for next to no real world gain.


The problem with self driving vehicles today is understanding the data that they are already receiving.

Adding more ways to get far more data is not going to solve anything.

 

[emmz0r]

This could work with dedicated taxis which is where people don't mind having a Christmas tree of sensors on it.
But on your personal car however...

 

[diplomat33]

Humans drive with two eyes and two ears.
Humans drive with a visual system of 20/60 or better. That is very crappy vision, and humans make it work.

I've always found this to be a poor argument. Sure, humans "make it work" but humans also fail a bunch of times and crash. Don't we want autonomous cars to be far better than humans? I do.

Computers should be able to do the same with the same equivalents. (Maybe a few extra if the computer cannot turn to do shoulder checks, but only a few extra)

That is easier said than done. Getting a computer to understand an image is harder than you might think. It is not obvious like it is with humans.

 

[Daniel in SD]

Really not as good as it sounds.

Your looking at a camera system that is far more expensive then the car it goes on top of.

There is no point making a system that no one can afford to own.


Humans drive with two eyes and two ears.
Humans drive with a visual system of 20/60 or better. That is very crappy vision, and humans make it work.

Computers should be able to do the same with the same equivalents. (Maybe a few extra if the computer cannot turn to do shoulder checks, but only a few extra)

Anything more then that is just increasing the price for next to no real world gain.


The problem with self driving vehicles today is understanding the data that they are already receiving.

Adding more ways to get far more data is not going to solve anything.

Yeah, why don't those dumb engineers just make a human level artificial intelligence?

What's the point in making a 2 seat EV that costs over $100k? You'll never sell more than a couple thousand.

 

[Daniel in SD]

There is no point making a system that no one can afford to own.

What's the point of a phone that weighs 2 pounds and costs $9000?
Why not just make a 4oz phone that costs $100?

 

[alstoralset]

Humans drive with two eyes and two ears.
Humans drive with a visual system of 20/60 or better. That is very crappy vision, and humans make it work.

Nobody really knows, but it's quite possible that AI capable of camera-only FSD will necessarily be able to pass the Turing test.

 

[kevindamen]

The expectation that a self driving car will never crash is not realistic, and the idea that you can just add more sensors to fix every situation is not practical.

Eventually they will get good enought that crashing is extremely rair. (Only after many years of improvements after they are in wide use in the real world.)


Before then the best we can hope for is maybe 5 times less likely to crash then a human driver.

Let's give a few examples that have happened to me in real life.

1) I have had concrete fall off of an over pass and land on the front of the car.

2) I have had a deer try to jump over my car from the back left corner to the front right, only to end up dieing on front windshield.


I dont care what sensors you have, they are not going to work in all cases to stop all accidents...

 

[Daniel in SD]

The expectation that a self driving car will never crash is not realistic, and the idea that you can just add more sensors to fix every situation is not practical.

Eventually they will get good enought that crashing is extremely rair. (Only after many years of improvements after they are in wide use in the real world.)


Before then the best we can hope for is maybe 5 times less likely to crash then a human driver.

Let's give a few examples that have happened to me in real life.

1) I have had concrete fall off of an over pass and land on the front of the car.

2) I have had a deer try to jump over my car from the back left corner to the front right, only to end up dieing on front windshield.


I dont care what sensors you have, they are not going to work in all cases to stop all accidents...

I think Waymo would be pretty happy to achieve parity with a human! I think you're overestimating how far along the state of the art is.
A cursory look at the CA accident reports shows that they got in 24 accidents last year in CA with a safety driver in the car. That's an accident every 60k miles! In Waymo's defense they just seem to get rear ended a lot.

 

[brianman]

Sure, humans "make it work" but humans also fail a bunch of times and crash. Don't we want autonomous cars to be far better than humans?

I don't think that's a sensor problem.

 

[nepenthe]

I don't think that's a sensor problem.

I think what diplomat is trying to say is, current vision systems are in fact good enough for FSD, we just need sensors with micrometer resolution from a hectometer off. =p

 

[diplomat33]

I think what diplomat is trying to say is, current vision systems are in fact good enough for FSD, we just need sensors with micrometer resolution from a hectometer off. =p

Uh no. That is not at all what I am saying.

 

[DrDabbles]

Your looking at a camera system that is far more expensive then the car it goes on top of.

You're. And no we're not. Where did you get that from?

There is no point making a system that no one can afford to own.

No one? Then how is Waymo putting them on the road? Good thing Tesla didn't take that approach early on.

Humans drive with two eyes and two ears.
Humans drive with a visual system of 20/60 or better. That is very crappy vision, and humans make it work.

Humans drive with a massive number of sensors, eyes, ears, inbuilt balance system that detects minute changes in acceleration forces, oh, and a massive brain to process everything.

This is a fundamentally broken argument, and it's tired at this point.

Computers should be able to do the same with the same equivalents.

Let me know when you make an artificial brain. And obviously the point is to make these cars superhuman rather than just as good as human. So add on lidar and radar while you're at it.

Anything more then that is just increasing the price for next to no real world gain.

Saving lives and operating superior to humans isn't "no real world gain".

The problem with self driving vehicles today is understanding the data that they are already receiving.

Which is to say the problem with self driving vehicles today is that they don't work. Even though we see Waymo out there right now.

Adding more ways to get far more data is not going to solve anything.

So says you. Different types of data are useful for different components of driving, so to say they aren't going to solve anything is unbelievably shortsighted.

This could work with dedicated taxis which is where people don't mind having a Christmas tree of sensors on it.
But on your personal car however...

Technology shrinks. Lidar units are already much smaller than their rooftop unit.

 

[croman]

Stop with the logic and information @DrDabbles . Jeez, this is the internet. No place for that stuff. So 20th century. Now we have memes.

 

[malcolm]

Careers – Waymo

Big recruitment push

 

[DrDabbles]

I just want to add, whether someone "beats" Tesla to Level 5 autonomy or not doesn't matter in the slightest. Autonomy is a nice to have for Tesla, and it changes absolutely nothing about the fact that they are producing the best battery pack and drive system in the EV industry. Watching GM's EV day was very cathartic for me, because it basically was a full day of "we're copying everything Tesla is doing, but we won't say their name".

 

[diplomat33]

I just want to add, whether someone "beats" Tesla to Level 5 autonomy or not doesn't matter in the slightest. Autonomy is a nice to have for Tesla, and it changes absolutely nothing about the fact that they are producing the best battery pack and drive system in the EV industry. Watching GM's EV day was very cathartic for me, because it basically was a full day of "we're copying everything Tesla is doing, but we won't say their name".

Battery and powertrain tech is key. Certainly, having the best EV gives Tesla a huge advantage but I don't think we can totally dismiss the impact that autonomous driving would have. If autonomous driving was so meaningless, why is Tesla focused on it so much? And why is Waymo, Cruise, and so many other companies focused on it so much? The fact is that L5 autonomy, when it happens, will be a total paradigm shift worth trillions of dollars.

 

[DrDabbles]

Battery and powertrain tech is key. Certainly, having the best EV gives Tesla a huge advantage but I don't think we can totally dismiss the impact that autonomous driving would have. If autonomous driving was so meaningless, why is Tesla focused on it so much? And why is Waymo, Cruise, and so many other companies focused on it so much? The fact is that L5 autonomy, when it happens, will be a total paradigm shift worth trillions of dollars.

The majority of companies focusing on L4/5 right now are all commercial taxi services. Which is neat, but doesn't change the auto market significantly. We have taxis right now, and everybody still buys their own car. Removing the taxi driver out of that equation doesn't change it at all.

As for why Tesla is focusing on self driving capabilities? For the same reason everybody has been developing ADAS for two decades now. Occupant safety. And personally, I think they need to focus more on that angle than attempting to say they'll be delivering self driving any day now. Iteratively adding safety features while developing a path to higher levels of autonomy is a massive winning strategy IMO, that Tesla should be focusing no. What other vendor could OTA add a feature to perform AEB if you were about to run a red light or stop sign? Tesla could.

Even Intel/MobilEye has admitted that the first generation of FSD is going to be aimed at commercial endeavors rather than consumers.

 

[Daniel in SD]

The majority of companies focusing on L4/5 right now are all commercial taxi services. Which is neat, but doesn't change the auto market significantly. We have taxis right now, and everybody still buys their own car. Removing the taxi driver out of that equation doesn't change it at all.

It makes taxis cheaper which does change the equation. Tesla said just last year that they're going to stop selling cars once the robotaxi code is done by the end of this year.

As for why Tesla is focusing on self driving capabilities? For the same reason everybody has been developing ADAS for two decades now. Occupant safety. And personally, I think they need to focus more on that angle than attempting to say they'll be delivering self driving any day now. Iteratively adding safety features while developing a path to higher levels of autonomy is a massive winning strategy IMO, that Tesla should be focusing no. What other vendor could OTA add a feature to perform AEB if you were about to run a red light or stop sign? Tesla could.

I agree. Tesla has been adding safety features and I suspect that FSD will remain a marketing hyped cash grab for the foreseeable future.