Autonomous Car Progress

[Daniel in SD]

The reddit thread has "insiders" that say this may have to do with how some of these companies design their break system, which discourages you from honestly reporting your fatigue levels, because you would get suspended without pay. It's an interesting point. Some people (including in the comments in that article) make a big deal about having trained drivers versus Tesla just crowd sourcing it to owners, but are trained drivers that do this as a full time job really better in terms of safety (taking into account fatigue vs training)?

Companies certainly have a strong incentive to reduce collisions since they're liable for their employees. A Google search doesn't seem to produce a clear answer about whether company owned vehicles have a higher or lower accident rate than personal vehicles. Of course autonomous vehicle testing has the automation complacency issue and it's the entire job (not just driving to and from a job site or doing deliveries) so it's quite a bit different than regular driving. My feeling is that making autonomous vehicle testing safe is a huge challenge as the systems get more capable. In 2018 Cruise said they have two test drivers in every vehicle, not sure if they are still doing that.
Tesla is definitely trying to be safe with their testing. Unfortunately they also have to worry about customer satisfaction and I wonder if a safe testing program would be very unpopular. Though now that I think about it a little more they should try harder and kick some of the YouTubers off the beta.

 

[stopcrazypp]

Companies certainly have a strong incentive to reduce collisions since they're liable for their employees. A Google search doesn't seem to produce a clear answer about whether company owned vehicles have a higher or lower accident rate than personal vehicles. Of course autonomous vehicle testing has the automation complacency issue and it's the entire job (not just driving to and from a job site or doing deliveries) so it's quite a bit different than regular driving.

Yeah, the key thing is driving (at various levels of engagement) is all they do at a job like this, while for other company vehicles, there is typically a break in the driving where they do something else, even if driving makes up a large part of their jobs. I guess taxi drivers and "ride sharing" drivers would be most similar occupations (and environment too), but I doubt most people consider them to be good examples of safe drivers (although it may be interesting to see stats).

My feeling is that making autonomous vehicle testing safe is a huge challenge as the systems get more capable. In 2018 Cruise said they have two test drivers in every vehicle, not sure if they are still doing that.

I'm pretty sure they have switched to single drivers already for the cars that still have drivers.
Here's a picture from a 2020 article for example taken in San Francisco:

Cruise raises $2.75 billion from Walmart, others
They even got approved for driverless cars last year:
Cruise is now testing fully driverless cars in San Francisco

Tesla is definitely trying to be safe with their testing. Unfortunately they also have to worry about customer satisfaction and I wonder if a safe testing program would be very unpopular. Though now that I think about it a little more they should try harder and kick some of the YouTubers off the beta.

I think it's already unpopular that there is a safety score system, although I don't think necessarily there have been as much pushback against being kicked off the program for not paying attention (false positives excepted). Would be interesting to see how many people they expanded to so far. AFAIK they are still limiting to around 98 score.

 

[S4WRXTTCS]

They hit someone on a scooter not long ago, also under manual control according to Waymo.

Those operators can be quite aggressive; a Cruise almost hit me by failing to yield at a crosswalk where I was almost halfway across, I bet also under manual control. A Zoox yielded to me at the same crosswalk another time and I wasn’t even crossing yet

One thing that concerns me is human expectations of what autonomous vehicles will do.

Driving, biking, and walking in public is a social activity where make decisions based on expectations.

With an autonomous vehicles that expectation is they'll be passive, and they'll stop for me when I get into the cross walk.

But, if a human is in control that expectation is completely wrong and could get a person killed.

I'm convinced that we'll see pedestrian, scooter, bike incidents with autonomous vehicles because the behavior of those vehicles was unexpected. I'm sure some of the incidents will be the result of these individuals bullying autonomous vehicles.

A zoox is pretty much begging to be bullied.

 

[Bladerskb]

Oh look another car model (ET5) announced for 2022 with yet another high resolution lidar and only cost $40 that will sale in the tens of thousands alone a year among the other dozens. Oh wait it doesn’t matter because look at the 2022 Mercedes that cost 100k and still uses a low resolution. That is clearly what automakers are doing.

This trend will continue as I predicted year(s) ago that by 2025 50% of all new EVs will have a front facing lidar.

 

[Terminator857]

Oh look another car model (ET5) announced for 2022 with yet another high resolution lidar and only cost $40 that will sale in the tens of thousands alone a year among the other dozens. Oh wait it doesn’t matter because look at the 2022 Mercedes that cost 100k and still uses a low resolution. That is clearly what automakers are doing.

This trend will continue as I predicted year(s) ago that by 2025 50% of all new EVs will have a front facing lidar.

When will these lidar vehicles be in the U.S. and outperform FSD beta?

 

[Daniel in SD]

When will these lidar vehicles be in the U.S. and outperform FSD beta?

That's like asking when someone will release something that outperforms Smart Summon.
My prediction is that nothing publicly released will outperform FSD Beta for decades. No one is ever going to release anything like it until driverless operation is achieved and even then FSD Beta will work in places that driverless vehicles won't.

 

[EVNow]

My prediction is that nothing publicly released will outperform FSD Beta for decades. No one is ever going to release anything like it until driverless operation is achieved and even then FSD Beta will work in places that driverless vehicles won't.

So you think GM is lying about releasing UltraCruise next year ?

 

[EVNow]

Companies certainly have a strong incentive to reduce collisions since they're liable for their employees.

Nothing like owners putting their own cars on the table. Besides they use FSD for short duration at a time (avg is probably less than 30 minutes) vs 8 hours a day for safety drivers.

 

[Daniel in SD]

So you think GM is lying about releasing UltraCruise next year ?

It's not a lie if you believe it.
I just can't imagine they'll do it. And of course it will never outperform FSD Beta. For example, I doubt it will cross double yellows to go around double parked cars.

 

[EVNow]

For example, I doubt it will cross double yellows to go around double parked cars.

Does Waymo do it ? If it does, why not GM - purely from a road rules perspective.

Ofcourse it could be a dumbed down version like that Ford AP that silently disengages for slight bends.

 

[Daniel in SD]

Nothing like owners putting their own cars on the table. Besides they use FSD for short duration at a time (avg is probably less than 30 minutes) vs 8 hours a day for safety drivers.

I guess I'll keep harping on this. I just don't see an owner maintaining vigilance after their car does a 30 minute commute flawlessly for an entire year.
Maybe you could have the system intentionally make errors and kick people off if they don't catch them? I'm sure there are ways to make it safe.

 

[Terminator857]

That's like asking when someone will release something that outperforms Smart Summon.
My prediction is that nothing publicly released will outperform FSD Beta for decades. No one is ever going to release anything like it until driverless operation is achieved and even then FSD Beta will work in places that driverless vehicles won't.

Mercedes is releasing level 3 hopefully. Others have announced level 3. So instead of decades just a few years? How does GMs hands free cruise compare? Releasing a version for city streets in 2023.

GM's Ultra Cruise to Offer Hands-Free City Driving for MY-2023

Ultra Cruise is designed to enable hands-free driving in 95% of all driving scenarios, and will provide a a door-to-door hands-free driving experience.

www.fleetforward.com

Hopefully Tesla will have something much better by 2023, with time we will know.

 

[EVNow]

I guess I'll keep harping on this. I just don't see an owner maintaining vigilance after their car does a 30 minute commute flawlessly for an entire year.
Maybe you could have the system intentionally make errors and kick people off if they don't catch them? I'm sure there are ways to make it safe.

I wish that time would come soon ... but at that time we are really talking robotaxi level accuracy.

ps : Then there is also the news about what is happening to others on FSD. What keeps a lot of people attentive about AP now is news about other AP related accidents (apart from issues they may see themselfs). But I've to say - I see most of the drivers are less attentive when driving manually than when I drive on FSD Beta or perhaps even AP. I think the trick is to have your hands on the wheel all the time.

 

[Terminator857]

... I just don't see an owner maintaining vigilance after their car does a 30 minute commute flawlessly for an entire year. ...

I think camera and hands on the wheels nag will make it difficult to goof off. But if that doesn't work I vote for electric shock therapy. Everytime you doze off the car sends some juice your way.

 

[stopcrazypp]

Oh look another car model (ET5) announced for 2022 with yet another high resolution lidar and only cost $40 that will sale in the tens of thousands alone a year among the other dozens. Oh wait it doesn’t matter because look at the 2022 Mercedes that cost 100k and still uses a low resolution. That is clearly what automakers are doing.

This trend will continue as I predicted year(s) ago that by 2025 50% of all new EVs will have a front facing lidar.

What model $40 high resolution lidar is it using? I'm not about to fall for the same thing as your last post where even a cursory search shows the companies you linked are still using low resolution lidar in the actual models that are released, while you link a tech demo of a different more expensive lidar model made by the same company.

 

[Bladerskb]

Innoviz detects lane lines quite clearly and is one of the features of their perception software:

You should dig a bit deeper. The HAP is NOT inferior to HORIZON. Its superior in every way.
First of all that 0.05 for Horizon is angular precision not angular resolution, they are not the same thing.

Horizon
64 lines at 10 hertz (10fps)
90 m @ 10% reflectivity
81.7° (Horizontal) ×25.1° (Vertical)
240,000 pts/s
Not Automotive Grade

HAP
144 lines at 10 hertz (10fps)
150 m @ 10% reflectivity
120° (Horizontal) x 25° (Vertical)
450,000 pts/s
Automotive Grade


96 lines IS high resolution lidar... Remember SDC used to use Velodyne 64 line high resolution lidar and that was all they used. Google used only one 64 line lidar for a very long time. Infact today alot still do. Most of cruise prototypes were made up of 32 lines velodyne lidars. I could keep going. To call 96 line low resolution is misinformation.

This is wrong, both livox HAP (Xpeng P5) and huawei lidars (Arcfox As and other chinese models) are being mass produced today and there are several more like innoviz that are waiting on BMW IX. But you keep changing the goal post rather than admitting you were wrong.

This isn't what you said, nor implied, you said it would "no doubt would be the type Ford would use if they opted for it".
You keep moving the goal post, why not just admit you were wrong?

no they are not. when 10-15 automakers are doing one thing, focus on the one that isn't and claim THIS IS WHAT EVERYONE IS DOING.
Typical tesla fan.

For the record i don't believe nor is it true that the lack of lidar is the reason for the curve problem nor the solution.


So a future 2022 Mercedes is now the deciding factor of what Lidar automakers will use? You tesla fans never fail to amaze me.
So the $30-50k cars with 3,4,5 and 6 high resolution lidar don't matter.
The 10-15 automakers using High resolution lidars don't matter.
But if a 100k mercedes uses a low res lidar then thats what the automakers are still using.
The logic of a tesla fan in full display. The funny part is, there will be more Livox HAP lidars sold through xpeng p5 this year than the mytical mercedes L3 all next year and probably the year after that.

Infact there will be more lidars sold by each individual lidar companies: luminar on volvo cars and other chinese cars, innovusion on ET7 and other NIO models, Huawei, and Livox than the Scala 2 through mercedes.

But that wont stop you from pushing your false narratives that automakers are still using low resolution lidars, just like it doesn't stop tesla fans from portraying blue cruise as "the competition".

@stopcrazypp continues to be the biggest source of misinformation in this thread.

Key to spotting a Tesla fanatic is when they continue to spread misinformation despite it being proven to be balatntly false and constant attempt to revise history. Calling 96 line lidar low resolution when 32 line is regarded industry wide as high resolution. In fact almost all sdc were using 32 line lidars and many still do to this day. Same with the 64 line velodyne. And the entire industry regards it as high res. Then he says Livox HAP is inferior to Livox Horizon. Just pure rubbish misinformation.

So @stopcrazypp continues his campaign of spreading misinformation going on 5 years now.

 

[Daniel in SD]

I wish that time would come soon ... but at that time we are really talking robotaxi level accuracy.

Driving 100 hours without an error isn't even close to robotaxi performance.

 

[Terminator857]

Driving 100 hours without an error isn't even close to robotaxi performance.

I suppose it depends on what an error is and who is judging. Waymo has a much higher bar for imperfections.
For Elon the bar is much lower: roller coaster rides r us. Elon: If we didn't hit anything hard then we are good.

 

[Daniel in SD]

I suppose it depends on what an error is and who is judging. Waymo has a much higher bar for imperfections.
For Elon the bar is much lower: roller coaster rides r us. Elon: If we didn't hit anything hard then we are good.

My bar is median human performance (which is probably quite a bit better than average!).
The other issue is you can't extrapolate self-driving car performance like you can for human performance. If you rode along with a human driver for 100 hours you could probably determine quite well how good a driver they are. A self-driving car could stay perfectly centered in its lane, smoothly navigate turns, and appear to be flawless for 100 hours. Then it could encounter a situation that it's not programmed to handle and fail catastrophically.

 

[stopcrazypp]

@stopcrazypp continues to be the biggest source of misinformation in this thread.

Key to spotting a Tesla fanatic is when they continue to spread misinformation despite it being proven to be balatntly false and constant attempt to revise history. Calling 96 line lidar low resolution when 32 line is regarded industry wide as high resolution. In fact almost all sdc were using 32 line lidars and many still do to this day. Same with the 64 line velodyne. And the entire industry regards it as high res. Then he says Livox HAP is inferior to Livox Horizon. Just pure rubbish misinformation.

So @stopcrazypp continues his campaign of spreading misinformation going on 5 years now.

My original discussion was talking about lidar good enough to detect lane lines, that the consumer lidar today is not good enough to do that. You were the one who was misleading by mentioning 640 lines in 2022 (which none of the lidar models confirmed so far going into production can get close to even though you were obviously implying they were on the same order), mentioning a bunch of cars using lidars with far lower resolution that are unlikely to be able to reliably detect lane lines, and then now changing the subject to 96 line lidars still fitting the "high resolution lidar" definition and getting further and further away from the original point (which was about lidar that can reliably detect lane lines and thus aid in avoiding the situation of not being able to make curves, even accepting the likely false premise that perception was the core cause of those failures). Then adding name calling to boot and accusing me of goal post moving.
Quoting the discussion below for others to follow.

While some LIDAR sensors have enough resolution and provide enough data (using reflectivity) to detect lane lines, the kind of consumer LIDAR that are being used in massed produced passenger cars today (like the Valeo Scala used by Audi and no doubt would be the type Ford would use if they opted for it) simply does not have the resolution to reliably detect lane lines even if they were perfect and on a clear day with clean roads.

2017 called...they want their lidar back.

All cars going into production today and in the near future uses high resolution lidars, not the 4 line lidar of Scala 1. To understand the difference in resolution and spec, Luminar which will be going into several cars in 2022 has 640 lines.

Current Production Cars in 2021 with lidar:
Lucid Motors (High resolution)
Xpeng P5 (High resolution Livox lidars)
Huawei Arcfox As Hi (High resolution Huawei Lidars)

Cars releasing in 2022 with lidar (not a complete list):
Volvo (High Resolution Luminar lidar)
Nio ET7 (High resolution Innovusion lidar)
BMW IX (High resolution Innoviz lidar)

Cars releasing in 2023 with lidar (not a complete list):
GM Lyriq (High Resolution Ceptron lidar)

All the lidars are high resolution and can read lane lines, road markings, curbs, etc

Luminar Lidar output:

Innoviz Lidar output:

Innovusion Lidar output:
nio.com/cdn-static/mynio/videos/nad/nad-lidar-highway.mp4

Livox Lidar output:

Not sure how you are coming to this conclusion.