Autonomous Car Progress

[Doggydogworld]

I just don't buy the robotaxi replacing personally owned cars argument. I get the conveniences, and I get their value in a dense urban environment where car ownership is already a luxury and a hassle due to limited space for parking and higher costs of ownership. But, the huge thing being overlooked here is that you have to bring along everything you need at all times with you when using a robotaxi because you can never store things in one because it isn't yours.

For families with kids and/or pets, that already becomes a pretty terrible value. Lug around car seats and install them every time you get in a robotaxi. That's just never going to fly. I always have a dog crate in our minivan along with some hiking supplies, hiking sticks, etc, so that going on hikes on the weekend with my kids and dog is just a matter of getting all of us into our car and heading over to our favorite hiking spots. With a robotaxi, you gotta lug all that stuff along, get it into the taxi, then go wherever... and then what? You pay to have the car wait for you with your crate and car seats in the parking lot for several hours while you hike, then return home and remove all of that again?

I'm just not buying the argument that robotaxis will ever replace ownership of vehicles. It will make sense for a subset of society, but I just don't see anyway for it to be the only (or even dominant) mode of transportation.

There's also the issue of wait time vs. hop in and go. First world problems, but people spend a lot of their car budget on sillier stuff. I do see Robos taking out a lot of 2nd/3rd cars, though. That's probably Waymo's thinking. Tiered service plans like $399/month for 800 miles would be a good 2nd car replacement, especially since it's more flexible (e.g. one Waymo takes Kelsie to piano lessons while another picks Billy up from baseball practice after school on the other side of town).

Speaking of wait time, this local reporter took a Waymo ride with decidedly mixed results. She met a Waymo rep in a strip mall, but it scheduled the pickup in a neighborhood. So they went there, then it took 19 minutes to show up (originally said 1 minute). Then it drove past them and stopped in the cul de sac a half block away! ""Obviously there have been some hiccups,.." said the Product Manager they interviewed. Seriously? After three %#*! years? Good grief. Great technology but they're just completely clueless when it comes to the business end of things.

The reporter said they've had 47 accidents in 16 million Chandler-area miles, "almost all" the fault of the other party. She also said they've given 100k rides since October (half to JJRicks, ha). I'd not heard either stat before. The ride itself was uneventful, the reporter mentioned how cautious it drove.

 

[gearchruncher]

So they went there, then it took 19 minutes to show up (originally said 1 minute).

I asked my Tesla to come pick me up in 2017 and it still hasn't arrived even though Tesla originally said 1 year.

It will make sense for a subset of society, but I just don't see anyway for it to be the only (or even dominant) mode of transportation.

As wealthy Americans, we need to be careful of our bias. Yes, keeping stuff in the car is nice, but with the median household income at $68K, the average house would figure out a way to store stuff somewhere else if it saved them $10 a day. Other products will show up to fill this void like lockers at work. 60% of all vehicle trips in the USA are under 6 miles, not big road trips or even out go to hiking. I see no issue with non-owned autonomous cars becoming a dominant mode of transport. Public transport is already very effective the world over and has common issues (can't store stuff, small delays in arriving, can't customize it). That doesn't mean there won't also be other ownership models, kind of like how horses moved from transport to hobbies. I do think that liability needs to be worked out before private ownership works too, as I'll never own a vehicle without a steering wheel while also being liable for any harm it causes, but I also don't know what sane company would sell me a car, retain liability, but leave maintenance up to me. I think the most likely model is an "exclusive lease" where you don't own it (like a lease today) but the company dedicates that vehicle to you. You'll just pay large sums for this in direct and indirect ways (lack of vehicle efficiency means high lease costs, parking in public, needing parking at your house, charging infrastructure...)

And I say all of this as someone who considers cars and driving cars to be my primary hobby.

 

[diplomat33]

The reporter said they've had 47 accidents in 16 million Chandler-area miles, "almost all" the fault of the other party. She also said they've given 100k rides since October (half to JJRicks, ha). I'd not heard either stat before. The ride itself was uneventful, the reporter mentioned how cautious it drove.

I think it is 47 incidents in 6.1M Chandler area miles. Waymo released a study on the data that details the types of accidents, severity of accidents etc...

You can read the paper here:

https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf

 

[Doggydogworld]

I think it is 47 incidents in 6.1M Chandler area miles. Waymo released a study on the data that details the types of accidents, severity of accidents etc...

You can read the paper here:

https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf

Good catch, that's definitely where the 47 accidents came from. Although 47 in 6.1m sounds bad (1 accident per 130k miles), their threshold for accident is extremely low. For example a pedestrian bumping into the van while it was stopped.

I think the 16m miles is total miles in the Chandler area from 2016-present, whereas the 6.1m miles and 47 accidents only covers 21 months in 2019-20. So she confused two different time periods.

I see no issue with non-owned autonomous cars becoming a dominant mode of transport. Public transport is already very effective the world over and has common issues (can't store stuff, small delays in arriving, can't customize it).

Even in the US, some in city centers where parking is an expensive hassle eschew car ownership. They use mass transit and cab/Uber/Lime/bike/scooter/whatever. If Robotaxis cost half as much as Uber many more urbanites will go carless. I still don't see families In suburbs and less dense cities going completely carless, just reducing from 2-4 cars down to 1-2.

 

[diplomat33]

Good catch, that's definitely where the 47 accidents came from. Although 47 in 6.1m sounds bad (1 accident per 130k miles), their threshold for accident is extremely low. For example a pedestrian bumping into the van while it was stopped.

Yeah, the bar is very low. The report includes every "point of contact" (ie where an object came into contact with the Waymo car). Most of the incidents were minor.

The scary part is reading about the bad behavior of some human drivers. I think there was a drunk driver that drove down the road in the wrong direction and almost hit the Waymo head on. Another human driver ran a red light. Yet another human driver was speeding and rear ended the Waymo. Yikes! It is a reminder that AVs can drive correctly and follow traffic laws and still get into accidents because of the human drivers around them driving badly and breaking traffic laws.

I think the 16m miles is total miles in the Chandler area from 2016-present, whereas the 6.1m miles and 47 accidents only covers 21 months in 2019-20. So she confused two different time periods.

Yes, I think she confused the two periods.

 

[daktari]

I just don't buy the robotaxi replacing personally owned cars argument. I get the conveniences, and I get their value in a dense urban environment where car ownership is already a luxury and a hassle due to limited space for parking and higher costs of ownership. But, the huge thing being overlooked here is that you have to bring along everything you need at all times with you when using a robotaxi because you can never store things in one because it isn't yours.

For families with kids and/or pets, that already becomes a pretty terrible value. Lug around car seats and install them every time you get in a robotaxi. That's just never going to fly. I always have a dog crate in our minivan along with some hiking supplies, hiking sticks, etc, so that going on hikes on the weekend with my kids and dog is just a matter of getting all of us into our car and heading over to our favorite hiking spots. With a robotaxi, you gotta lug all that stuff along, get it into the taxi, then go wherever... and then what? You pay to have the car wait for you with your crate and car seats in the parking lot for several hours while you hike, then return home and remove all of that again?

I'm just not buying the argument that robotaxis will ever replace ownership of vehicles. It will make sense for a subset of society, but I just don't see anyway for it to be the only (or even dominant) mode of transportation.

Agree.
Where are all the robotaxis going to park if one need swift service? In peoples driveway? Or shall they roll around, always on the move, ready?
Here in Oslo the city is now swamped with electric scooters that showed up last year, 5 companies rent them out. People park these everywhere, just throw them away at the sidewalk and everywhere. But that is the convenient thing with them, scoot along to the destination, drop it. Sadly the city looks littered.
I am worried about this aspect for the robotaxies also.

A sidenote: what kind of robotaxi would you call? A special made super comfort box with soft and big seats, perfect for relaxation and conversation, cool ambient lights etc, OR a low, slightly cramped sport sedan with stiff suspension, small doors and noise levels that make raising the voice a necessity to hear eachother? Depends on the occasion of course; the soft cruiser for a dinner out, the roomy and rugged transporter for a robotrip to the ski resort and so on. But when would anyone want a sporty "drivers car" as a robotaxi?

 

[PhantomX]

Yeah, the bar is very low. The report includes every "point of contact" (ie where an object came into contact with the Waymo car). Most of the incidents were minor.

The scary part is reading about the bad behavior of some human drivers. I think there was a drunk driver that drove down the road in the wrong direction and almost hit the Waymo head on. Another human driver ran a red light. Yet another human driver was speeding and rear ended the Waymo. Yikes! It is a reminder that AVs can drive correctly and follow traffic laws and still get into accidents because of the human drivers around them driving badly and breaking traffic laws.


Yes, I think she confused the two periods.

Just curious, what's the expectation of a self driving car when it comes to evasive maneuvers? Like you mentioned, there are times when people drive badly where the self driving car needs to react to stay safe and not necessary follow the rules of road. Has any company working on autonomous vehicle discussed this?

 

[dgatwood]

As wealthy Americans, we need to be careful of our bias. Yes, keeping stuff in the car is nice, but with the median household income at $68K, the average house would figure out a way to store stuff somewhere else if it saved them $10 a day.

It's challenging when you're using a car for things other than just driving to work. For example, I drive to music rehearsals after work, which involves carrying a trombone, music, a stand, and a mute bag. It's nontrivial to carry all of that stuff on top of the stuff for work, much less to find a place at work to store it, or find a place at the rehearsal space to store the stuff from work. And if I went back home in between, it would add an extra hour to my evening commute.

And when going out shopping, you usually shop at more than one place. You're not going to want to haul all of that stuff out of your car and store it at another place of business, realistically, and you're going to be shopping for up to a couple of hours at each place.

So it's challenging even in an urban area. But in rural areas, the amount of wait time for a vehicle would be high enough that sharing cars would be completely impractical. Imagine waiting twenty minutes for a car every time you need to go somewhere. At some point, you might as well walk.

Other products will show up to fill this void like lockers at work.

Maybe, but at what cost? At some point, it's cheaper just to keep the car longer — particularly when you're talking about Bay Area square footage prices.

 

[stopcrazypp]

As wealthy Americans, we need to be careful of our bias. Yes, keeping stuff in the car is nice, but with the median household income at $68K, the average house would figure out a way to store stuff somewhere else if it saved them $10 a day.

I think you underestimate the ability of the average American to make unwise financial decisions just to own a car they like. That's why there's all those subprime auto loans.

 

[stopcrazypp]

That doesn't matter especially when we are talking about personally owned robot-taxi. If waymo were selling their car with their system today for 35k, it doesn't matter what the car cost without their system.

You pay $35k and you get the entire system (hardware and software).

In the same way the current costs are:

Tesla's Model S system (8 cameras and one radar, 144 Tops Tesla computer) at the moment cost +$80k.

Nio's ET7 Model S competitor (11 cameras, 5 radars, 1 Hi-res lidars, 1000 Tops Nvidia Computer) at the moment cost ~$69k.

Huawei's Arcfox Alpha S system (3 Hi-res Lidars, 12 cameras, 6 radars & 400 TOPS Huawei 400 TOPs computer) at the moment cost $60-66k.

Tesla's Model 3 system (8 cameras and one radar, 144 Tops Tesla computer) at the moment cost $50k.

Xpeng's P5 system (12 cameras, 6 radars, 2 Hi-res lidars, ? Tops Nvidia computer) at the moment cost $35k.

As far as I kind find, none of the non-Tesla options you list actually exist yet. They are all in preorder status, and it's unclear which option and how much it would be to buy one with the actual sensors. As I said, let's see when the cars are actually released.

Interesting because they have such low resolution compared to others. But Its not surprising that their only partners are traditional automakers who probably spec'ed this out 3-4 years ago and are allergic to change so they are stuck with it. Other OEMs lidars that have orders of magnitude more resolution for less money. Luminar who has one of the best resolution on the automotive market and cost $500. Others like robosense that cost $200, Huawei is trying to get their lidar down to $200 which means they are hovering around that number. Livox is most likely around $100-$250 aswell. Innoviz one is just under $1,000.

To put that into perceptive, these lidars used to cost around 75k-100k. So we are talking about orders of magnitude improvement on cost and quality

The ones that cost 75k-100k are the ones that are mounted on the roof and can get a whole 360 degree view. The current ones that may cost in the hundreds (made in high volume) have a much narrower FOV and will require multiple to get the same view. I should also point out they are aiming at L2+ or L3 with these, not at L4/L5 yet; I think the L4/L5 players from the previous comment are arguing sensor suite comparable to theirs (which gets a whole 360 degree view) would be required for L4/L5.

 

[diplomat33]

Just curious, what's the expectation of a self driving car when it comes to evasive maneuvers? Like you mentioned, there are times when people drive badly where the self driving car needs to react to stay safe and not necessary follow the rules of road. Has any company working on autonomous vehicle discussed this?

I think self-driving cars could be expected to be good at defensive driving to try to avoid accidents as much as possible and engage in evasive maneuvers within reason as long as they don't cause another accident. Some accidents will be unavoidable no matter what and you don't want to engage in reckless driving just to try to avoid an accident. Considering the safety data that Waymo has, I would expect them to be working on improving their defensive driving to try to avoid some of those accidents caused by other drivers. Self-driving cars with their long range 360 degree sensors would be in a unique position to hopefully detect a bad driver with enough advance notice to try to avoid a potential accident.

 

[gearchruncher]

As far as I kind find, none of the non-Tesla options you list actually exist yet. They are all in preorder status, and it's unclear which option and how much it would be to buy one with the actual sensors. As I said, let's see when the cars are actually released.

Isn't Tesla FSD in preorder status also? As Tesla has taught us, shipping hardware isn't really worth anything if you don't have the software also. I mean, it's turning out that Tesla shipped hardware they aren't even going to use when they eventually, someday, we promise, actually ship some level of FSD.

 

[stopcrazypp]

Isn't Tesla FSD in preorder status also? As Tesla has taught us, shipping hardware isn't really worth anything if you don't have the software also. I mean, it's turning out that Tesla shipped hardware they aren't even going to use when they eventually, someday, we promise, actually ship some level of FSD.

Given this discussion is about hardware, this is talking about preorder for hardware first. You need the hardware first to even begin talking about software.

Software is a whole other story, as some of the cars listed are from companies whose production software is not even as advanced as Tesla's L2 (doesn't handle curves as well, doesn't handle local roads and traffic lights like Tesla does currently). Their current released software is fighting Tesla's NOA/EAP (and to be fair does do better than NOA in some regards, especially given they chose a better mapping service in China; Tesla is using Baidu, which is inferior). They operate under a similar model as Tesla, meaning the software is continually going to be under development, it's not like traditional automakers where most of the functionality will be there at launch.

 

[gearchruncher]

Given this discussion is about hardware, this is talking about preorder for hardware first. You need the hardware first to even begin talking about software.

Most good architectures are a combination of HW and SW. It's not like Tesla is showing us a good knowledgeable lead on hardware- they have told us HW1, HW2, HW2.5, and HW3 would all be able to self drive much more than any one has yet, and they had radar for 7 years now that apparently they don't need. So why trust them that LIDAR or any other specific hardware isn't needed?

It's just very hard to say that Tesla has for sure shipped hardware needed for autonomy while everyone else is "preorder". We have no idea what is needed, nobody has shown it for privately owned cars. All we know is the only systems out there doing L4 have LIDAR, so in that case Tesla hasn't put the hardware out their either. They've only delivered hardware if you are sure they won't need LIDAR or more cameras or more radar or XXX, and the only way we'll know that is if they actually deliver the SW.

 

[Bladerskb]

As far as I kind find, none of the non-Tesla options you list actually exist yet. They are all in preorder status, and it's unclear which option and how much it would be to buy one with the actual sensors. As I said, let's see when the cars are actually released.

As far as i know, the NIO ET7 comes standard with all sensors including the lidar.
Its also confirmed that the $60 & $66k huawei arcfox alphas s has the all the hardware and sensors.
These are not cars years into the future. These are cars that are acouple of months away.

The ones that cost 75k-100k are the ones that are mounted on the roof and can get a whole 360 degree view. The current ones that may cost in the hundreds (made in high volume) have a much narrower FOV and will require multiple to get the same view.

If you compare the quality, reliability and resolution of lidars today, its night and day.
For example lets look at the 75k lidar Velodyne HDL 64E, remember that this lidar needed constant maintenance and there were reports it broke down every month, this is in contrast to production lidars that are specc'ed for 50,000+ hours (decades of driving) and tested for all conditions.

Velodyne HDL 64E - $75k

Key Features:​

• 64 lines
• 50m (10% reflectivity), 120m (80% reflectivity) range
• 360° Horizontal FOV
• 26.9° Vertical FOV
• 0.08° angular resolution (azimuth)
• <2cm accuracy
• ~0.4° Vertical Resolution

https://hypertech.co.il/wp-content/uploads/2015/12/HDL-64E-Data-Sheet.pdf

Luminar Iris - $500

Key Features:​

• 640 lines
• 500m max range
• 250m at <10% reflectivity
• 120° Horizontal FOV
• 30° Vertical FOV
• 0.07° horizontal resolution
• 1cm accuracy
• 0.03° Vertical Resolution
• Dust & Water Ingress, Vibration & Shock certified

I should also point out they are aiming at L2+ or L3 with these, not at L4/L5 yet; I think the L4/L5 players from the previous comment are arguing sensor suite comparable to theirs (which gets a whole 360 degree view) would be required for L4/L5.

With lidars like Luminar that have 120 FOV. You only need three for a 360 degree FOV. That's $1.5k versus $75k while being orders of magnitude better in every way. Also luminar has dozens of robotaxi partners, for example Mobileye is using 3x luminar for their robot-taxi and it cost Mobileye $1,000 per due to low volume. Livox is partnering with AutoX.

 

[heltok]

I thought this was a good overview of the codebase of OpenPilot from comma.ai:

Looks a lot cleaner than the legacy industry I used to work for...

 

[Doggydogworld]

The ones that cost 75k-100k are the ones that are mounted on the roof and can get a whole 360 degree view. The current ones that may cost in the hundreds (made in high volume) have a much narrower FOV and will require multiple to get the same view.

Even the 360 degree spinners don't cost 75k any more. Ouster's OS2-64 is $20k list, but they'll do volume pricing for as few as 5 units. Chinese companies like Hesai also undercut industry pioneer Velodyne. But specs on solid state/MEMS/etc. units are catching up rapidly, and they're cheap enough to put one on each corner.

 

[dgatwood]

With lidars like Luminar that have 120 FOV. You only need three for a 360 degree FOV.

Only if you mount them on the roof in a tight cluster. Otherwise, you're probably going to have to use four so that they overlap. And really, you'll probably have to do that anyway, because perfect alignment is a fantasy.

 

[mark95476]

Is there a reason why there's no examples from major OEMs? It's impossible to not notice the huge sensor on the roof of Waymo, Nuro, Cruise, ..etc... is missing from consumer vehicles.

Volvo has been around, but they're small, and got swallowed-up by Geely a while ago.
NIO probably has a bright-future, but do they even sell anything outside of China currently?

This Volvo design (not shipping) is the only one I've seen with a lidar "bulge". All shipping cars hide their lidars in the grill, behind the glass and/or in headlight/taillight assemblies.

View attachment 659491

Nio ET7

 

[mark95476]

I can see cheap and very abundant robotaxis pushing ownership to a subset of society. It'll help fill the void that poor public transportation has left in North America. I used to ride share with Uber/Lyft a lot even though taxis/buses/shuttles/trains have been around for years just because they were easily accessible and super convenient.

That said, I'll probably always own a vehicle. I'm all for better public transportation, pedestrian and cycling infrastructure so I don't have to drive.

I just don't buy the robotaxi replacing personally owned cars argument. I get the conveniences, and I get their value in a dense urban environment where car ownership is already a luxury and a hassle due to limited space for parking and higher costs of ownership. But, the huge thing being overlooked here is that you have to bring along everything you need at all times with you when using a robotaxi because you can never store things in one because it isn't yours.

For families with kids and/or pets, that already becomes a pretty terrible value. Lug around car seats and install them every time you get in a robotaxi. That's just never going to fly. I always have a dog crate in our minivan along with some hiking supplies, hiking sticks, etc, so that going on hikes on the weekend with my kids and dog is just a matter of getting all of us into our car and heading over to our favorite hiking spots. With a robotaxi, you gotta lug all that stuff along, get it into the taxi, then go wherever... and then what? You pay to have the car wait for you with your crate and car seats in the parking lot for several hours while you hike, then return home and remove all of that again?

I'm just not buying the argument that robotaxis will ever replace ownership of vehicles. It will make sense for a subset of society, but I just don't see anyway for it to be the only (or even dominant) mode of transportation.