-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Waymo's Fifth Generation Lidars, Radars and Cameras
- Thread starter Bladerskb
- Start date
Daniel in SD
(supervised)
Hi DrDabbles,
Who is the current commercial customer for these Waymo units???
Waymo can certainly put its prototypes and engineering models on the road that are too expensive
to sell commercially....
Just asking?
Shawn
Who is the current commercial customer for these Waymo units???
Waymo can certainly put its prototypes and engineering models on the road that are too expensive
to sell commercially....
Just asking?
Shawn
willow_hiller
Well-Known Member
The last estimate I heard from The Verge for the cost of compute and sensors in a Waymo vehicle was $400,000.
If the first post in this thread is accurate, that means they might have brought it down to $200,000 per vehicle. It's a good start, but still too expensive for anyone but a taxi business to own.
For a taxi business it may be feasible; vehicle cost is always a one-time payment, but the taxi driver salary is paid on a recurring basis. Assuming a taxi costs $30,000 and taking the average monthly salary of a US taxi driver from Indeed ($2,500), a $200,000 robo-taxi would start to pay dividends after 68 months.
Hi Willow-Hiller,
Thank you for the pricing information.
The technology is awesome...
Taxis will be a great place for it.
I hope to see their continuous
improvements and cost reductions so that someone other
than taxis and governments are able to afford it...
Shawn
Thank you for the pricing information.
The technology is awesome...
Taxis will be a great place for it.
I hope to see their continuous
improvements and cost reductions so that someone other
than taxis and governments are able to afford it...
Shawn
Perfect algorithm, perfect sensors, perfect and consistent AI = ?
Gridlock.
The 'simple' part that is missing is trusted and secure inter-vehicle comminucation. Not only would that increase the effective resolution of relatively simple sensors, but it is essential anyway to facilitate negotiation where there is alternate single lane traffic flow, priority for emergency vehicles and in super high congestion levels with very slow moving traffic and small spaces between vehicles.
Gridlock.
The 'simple' part that is missing is trusted and secure inter-vehicle comminucation. Not only would that increase the effective resolution of relatively simple sensors, but it is essential anyway to facilitate negotiation where there is alternate single lane traffic flow, priority for emergency vehicles and in super high congestion levels with very slow moving traffic and small spaces between vehicles.
Last edited:
That was a guess from a The Verge journalist without knowledge of the industry who talked about rumors of what a SDC currently cost. That has absolutely NOTHING to do with Waymo who has mature systems and are even using in-house hardware. Although off the shelf hardware are also cheap compared to before.
Waymo cars cost way less.
For example in 2012, their cars cost $125k and they used Velodyne's $75k dome lidar.
Then they developed their own lidar and in 2017 disclosed that the lidar that replaced the Velodyne lidar cost 90% less.
This means their 360 dome lidar cost $7,500.
In 2018 Waymo started selling their surround perimeter lidar (honeycomb) for under $5k.
There are 5x of those on their cars. So if they are selling it for under $5k, it means the actual cost to make would be able 70% or less of the retail cost. So that would put the sensors at $3,500 per lidar.
Waymo also uses their own in-house built TPU v3, GPU and Intel's CPU. Knowing how cheap chips are and since they are also using their own chip. I can guarantee their entire compute probably cost ~$5,000.
Automotive cameras at scale cost just $10 dollars. Even the new sony camera with 7M camera (IMX324) that supports Mobileye's EyeQ4 and EyeQ5 compared to Tesla's 1M camera is cheap at mass scale. However some SDC companies development cars use Blackfly GigE cameras which cost ~$800. But lets say Waymo is using the most sophisticated camera and lens out there and they are able to reduce the cost to a conservative $150 which they usually do. Their new system has 29 cameras.
Waymo has 4x radar systems. Radars are really really cheap and are around ~$100 at automotive scale. However Waymo said they created a whole new type of radar(4D imaging radar). So let's say each system cost $500.
So total:
- 1x Dome = $7,500 (as of 2017)
- 5x Honeycomb = $3,500 x 5 = $17,500 (as of 2019)
- 29x Cameras = $150 x 29 = $4350
- 4x Radar systems = $500 x 4 = $2,000
- TPU, GPU & CPU = $5,000
Total = $39,850
This is without even taking into account their 2020 reduction of the cost by half.
Waymo to start selling standalone LiDAR sensors – TechCrunch
Lidar - Laser Bear Honeycomb – Waymo
Last edited:
Well, their technology package has gained serious interest from all kinds of places from robotics to transit and transport industry. Now that it's proven to work in a Level 4 situation, it becomes more viable for commercial customers. As for price, the assumptions you see tend to be from complete guesses from many years ago. Since Waymo isn't building their own cars now, and they're making their own lidar units, costs have been decreased dramatically.
And this doesn't take into account any of the dramatic reductions that Waymo has reported over the past couple years without publishing hard numbers.
These numbers are completely made up. If you can't imagine building a Waymo system for under $200k, then I don't think you're paying much attention to the industry announcements about custom fabbing silicon and the rapid decline in lidar unit costs.
willow_hiller
Well-Known Member
They may be made up, but not by me. By some unnamed experts quoted often by The Verge. This particular article was about Cruise, but I first encountered the stat in their video about Waymo Exclusive look at Cruise’s first driverless car without a steering wheel or pedals
Just did some searching around, and they have been quoting the same unnamed experts since around 2017...
But are these individual components you've found on the web the same tech used in the original post vehicle?
I'd like to see the sales price of a Waymo vehicle, not a few parts here and there to start to be able to compare. The only estimates I've seen say that Waymo cars are vastly more expensive than Teslas.
The Waymo system is ugly, inefficient aerodynamically and not using hive learning to improve the fleet intelligence - not to mention being an ICE car. They seem to be focusing on trying to make a single unit very very good. While admirable, real world learning seems likely to produce artificial intelligence capable of handling dynamic situations when it is based on the input of hundreds of thousands of cars, eventually millions and tens of millions of miles driven and soon even more miles.
It's never wise to doubt Google, but, this different approach from Tesla is going to produce different results. Maybe they will be better. Maybe not. Without knowing the true costs of such vehicles and seeing them operate in very substantial numbers in the real world, predictions about which system is better, farther advanced, etc. seems like abject speculation to me.
I'd like to see the sales price of a Waymo vehicle, not a few parts here and there to start to be able to compare. The only estimates I've seen say that Waymo cars are vastly more expensive than Teslas.
The Waymo system is ugly, inefficient aerodynamically and not using hive learning to improve the fleet intelligence - not to mention being an ICE car. They seem to be focusing on trying to make a single unit very very good. While admirable, real world learning seems likely to produce artificial intelligence capable of handling dynamic situations when it is based on the input of hundreds of thousands of cars, eventually millions and tens of millions of miles driven and soon even more miles.
It's never wise to doubt Google, but, this different approach from Tesla is going to produce different results. Maybe they will be better. Maybe not. Without knowing the true costs of such vehicles and seeing them operate in very substantial numbers in the real world, predictions about which system is better, farther advanced, etc. seems like abject speculation to me.
That was a guess from a The Verge journalist without knowledge of the industry who talked about rumors of what a SDC currently cost. That has absolutely NOTHING to do with Waymo who has mature systems and are even using in-house hardware. Although off the shelf hardware are also cheap compared to before.
Waymo cars cost way less.
For example in 2012, their cars cost $125k and they used Velodyne's $75k dome lidar.
Then they developed their own lidar and in 2017 disclosed that the lidar that replaced the Velodyne lidar cost 90% less.
This means their 360 dome lidar cost $7,500.
In 2018 Waymo started selling their surround perimeter lidar (honeycomb) for under $5k.
There are 5x of those on their cars. So if they are selling it for under $5k, it means the actual cost to make would be able 70% or less of the retail cost. So that would put the sensors at $3,500 per lidar.
Waymo also uses their own in-house built TPU v3, GPU and Intel's CPU. Knowing how cheap chips are and since they are also using their own chip. I can guarantee their entire compute probably cost ~$5,000.
Automotive cameras at scale cost just $10 dollars. Even the new sony camera with 7M camera (IMX324) that supports Mobileye's EyeQ4 and EyeQ5 compared to Tesla's 1M camera is cheap at mass scale. However some SDC companies development cars use Blackfly GigE cameras which cost ~$800. But lets say Waymo is using the most sophisticated camera and lens out there and they are able to reduce the cost to a conservative $150 which they usually do. Their new system has 29 cameras.
Waymo has 4x radar systems. Radars are really really cheap and are around ~$100 at automotive scale. However Waymo said they created a whole new type of radar(4D imaging radar). So let's say each system cost $500.
So total:
- 1x Dome = $7,500 (as of 2017)
- 5x Honeycomb = $3,500 x 5 = $17,500 (as of 2019)
- 29x Cameras = $150 x 29 = $4350
- 4x Radar systems = $500 x 4 = $2,000
- TPU, GPU & CPU = $5,000
Total = $39,850
This is without even taking into account their 2020 reduction of the cost by half.
Waymo to start selling standalone LiDAR sensors – TechCrunch
Lidar - Laser Bear Honeycomb – Waymo
willow_hiller
Well-Known Member
I am really curious what Alphabet's plan for a business model is at this point.
If it is "Sell complete self-driving vans to businesses" surely they would have named a target price per vehicle and started taking preorders by now.
If it is "Keep self-driving tech to ourselves and hold a monopoly over robotaxis," why are they selling the components to potential competitors?
If it is "Sell complete self-driving vans to businesses" surely they would have named a target price per vehicle and started taking preorders by now.
If it is "Keep self-driving tech to ourselves and hold a monopoly over robotaxis," why are they selling the components to potential competitors?
Daniel in SD
(supervised)
"Waymo’s mission is to make it safe and easy for people and things to get where they’re going. The Waymo Driver can improve the world's access to mobility while saving thousands of lives now lost to traffic crashes."
I think eventually they're planning on licensing their technology. The first applications will be their own robotaxi fleet and trucking fleet.
As far as I know they have no plans to build their own vehicles.
Samsung sells components to Apple. It's not unusual to sell components to competitors.
Robo-taxi does not need rest and work 40 hours a week so it covers 2 or 3 drivers.
Beprimeau
Member
I agree. I would guess 90% of accidents are inattention, 5% road conditions, and 5% visibility. So just using eyes and ears could eliminate 90% of accidents, and Lidar and radar will only help with 5% or less.
if one can eliminate that 5% chance of getting an accident, it will be priceless,
That is when inattentive humans don't use their eyes and ears.
When humans are attended to their driving and follow rules and common sense in the traffic, two eyes works very close to perfect.
8 cameras pointing in all relevant directions, connected to a permanently attended computer, works 100 closer to perfect.
Soda Popinski
Member
An analyst on Tesla Daily didn't think the Tesla Network would be feasible without a partner due to a chicken and egg issue, even if they get Robotaxies working. If there aren't enough Tesla robotaxies available, nobody is going to rely on them, like you can with Uber/Lyft where you can get a driver in a few minutes. I don't think I'd be willing to wait 15 or 20 minutes for a robotaxi, when I can get an Uber in less than 5.
I'd imagine Waymo with even fewer cars available will have the same issue.
I'd imagine Waymo with even fewer cars available will have the same issue.
Daniel in SD
(supervised)
Tesla has said they will stop selling cars to consumers after the robotaxi code is done. They could easily reach Uber scale in less than a year and of course they would roll out in individual markets not the entire country at once. Worrying about anything other than their ability to actually make a robotaxi is silly IMHO.
Similar threads
- Article
- Article