Cadillac Super Cruise vs Tesla Autopilot

[ewoodrick]

Why can't Autopilot see that this is a semi truck? Unfortunately there are things that humans cannot do yet, like make a computer vision systems reliable enough for autonomous driving.
Now there's a good chance that Autopilot calculated that there was some probability this was a semi truck and also that there was some probability it was an overpass. If your LIDAR map says that there is no overpass there then that makes it more likely to recognize it as a semi truck.
View attachment 510930

I NEVER said that autopilot was perfect today.

You are right on if a LIDAR map says that it isn't there, you may could react. BUT, what if it is a new bridge?
And the Caddies LIDAR mapping is only a VERY small amount of roads, 4+ lane limited access, divided, if I remember correctly. And really isn't too feasible on surface streets, where roads change way too much.

And honestly, heavenly powers aren't make humans that are reliable enough for driving either.

 

[ewoodrick]

Here is a slide from a presentation from about 3 years ago that compares Tesla's AP2 to Ford's Autonomous Vehicle:


We can see that if Tesla can develop excellent computer vision, then Tesla's approach could lead to full autonomous driving at a lower cost and with better EV range, two big advantages for Tesla. Of course, getting that camera vision is the hard part. Other companies that use lidar have a more expensive solution that cuts down on range but it is a faster path to FSD and achieves higher safety.

Yes, the Ford solution is doing a great job.

Oh wait, where is the Ford solution?

Maybe it's not a faster path.

 

[diplomat33]

Yes, the Ford solution is doing a great job.

Oh wait, where is the Ford solution?

Maybe it's not a faster path.

Yes, it is a faster approach. Companies like Waymo and Cruise already have L4 autonomy now. Waymo has driverless robotaxis taking passengers on public roads today. Ford is just a stand-in for all the lidar approaches since Waymo, Cruise, etc are all using basically the same sensor suite.

 

[Daniel in SD]

You are right on if a LIDAR map says that it isn't there, you may could react. BUT, what if it is a new bridge?

Then you have a higher chance of phantom braking at that spot. I chuckle at the idea of a new bridge over an existing road. That probably happens a few times a decade in California.

And the Caddies LIDAR mapping is only a VERY small amount of roads, 4+ lane limited access, divided, if I remember correctly. And really isn't too feasible on surface streets, where roads change way too much.

I think it's completely feasible if you have a fleet of cars that can update the maps. With a small fleet of cars Google maps the entire US every few years.

 

[ewoodrick]

Then you have a higher chance of phantom braking at that spot. I chuckle at the idea of a new bridge over an existing road. That probably happens a few times a decade in California.

I think it's completely feasible if you have a fleet of cars that can update the maps. With a small fleet of cars Google maps the entire US every few years.

Probably a half dozen new bridges within 20 miles of me. A dozen significantly modified. I've got a bunch of new roundabouts and lots of road widening.
What happens if there happens to be a deer crossing the Interstate near a bridge? Should a car ignore it?

Every few years won't cut it.

I had a tree down on a road yesterday and construction crews were out. Detailed road maps won't help that. "But the vision system can see that" Yep, exactly, and that's why if the vision system is working, it really doesn't need detailed maps. I've got some roads on which they've moved the road over 2 feet for construction.

About the only thing that you can rely on with detailed maps is that they will be wrong. And the more detailed they are, the more wrong they will be. Detailed maps can definitely help you staying in-between the lines. But that's actually one of the easiest things to do.

 

[diplomat33]

I think it's completely feasible if you have a fleet of cars that can update the maps. With a small fleet of cars Google maps the entire US every few years.

Yes, and Mobileye is doing this now. They call it "Roadbook":

 

[diplomat33]

Yep, exactly, and that's why if the vision system is working, it really doesn't need detailed maps.

That's easier said than done. Nobody has a vision system that works well enough yet.

 

[ewoodrick]

That's easier said than done. Nobody has a vision system that works well enough yet.

And no one has a LIDAR that's working well either. Your point?

 

[diplomat33]

And no one has a LIDAR that's working well either. Your point?

Actually, lidar does work well enough to do FSD. Waymo and Cruise have proven that you can do reliable, safe, full autonomous driving with camera + radar + lidar + HD maps. So you can definitely do safe FSD with lidar. It's still an open question whether you can do safe FSD without lidar.

 

[nepenthe]

Actually, lidar does work well enough to do FSD. Waymo and Cruise have proven that you can do reliable, safe, full autonomous driving with camera + radar + lidar + HD maps. So you can definitely do safe FSD with lidar. It's still an open question whether you can do safe FSD without lidar.

To be fair, Cruise doesn't have a public offering yet, right? As well, Waymo is in very limited public release - both of which are operating in very small and specific geographical locations. Waymo being in one the probably easiest environments to operate in. Since neither are in wide release, I'm not sure we can say, either approach is "done" enough. Perhaps if Tesla just focused on their sensor suite for driving only in Phoenix, Arizona they also would be at a similar functional level for the area.

 

[diplomat33]

To be fair, Cruise doesn't have a public offering yet, right? As well, Waymo is in very limited public release - both of which are operating in very small and specific geographical locations. Waymo being in one the probably easiest environments to operate in. Since neither are in wide release, I'm not sure we can say, either approach is "done" enough. Perhaps if Tesla just focused on their sensor suite for driving only in Phoenix, Arizona they also would be at a similar functional level for the area.

Waymo and Cruise are not completely done obviously. But I think they have proven that their systems do work. Waymo is accepting passengers without a driver and has done millions of autonomous miles. Cruise autonomous cars also have millions of autonomous miles and can drive thousands of miles on difficult San Francisco streets without even a fender bender. That certainly proves to me that Waymo and Cruise have a very good autonomous driving system. Tesla has not proven that they can do that yet. Yes, maybe if Tesla had focused on just doing FSD in one area, they would have something similar but they did not do that. I can only compare what I see Waymo and Cruise show us versus what Tesla has publicly shown us. When Tesla releases "City NOA" then I can better compare what Tesla has put out with what Waymo and Cruise have put out.

Bottom line: Waymo/Cruise have demonstrated L4 robotaxis driving autonomously on difficult city streets. So far, Tesla has just given us a L2 driver assist. So right now, as it stands, Waymo/Cruise with lidar have achieved FSD and Tesla without lidar has not achieved FSD yet. I am not saying that Tesla will never do FSD. But that is the state of things right now.

 

[ewoodrick]

Waymo and Cruise are not completely done obviously. But I think they have proven that their systems do work. Waymo is accepting passengers without a driver and has done millions of autonomous miles. Cruise autonomous cars also have millions of autonomous miles and can drive thousands of miles on difficult San Francisco streets without even a fender bender. That certainly proves to me that Waymo and Cruise have a very good autonomous driving system. Tesla has not proven that they can do that yet. Yes, maybe if Tesla had focused on just doing FSD in one area, they would have something similar but they did not do that. I can only compare what I see Waymo and Cruise show us versus what Tesla has publicly shown us. When Tesla releases "City NOA" then I can better compare what Tesla has put out with what Waymo and Cruise have put out.

Bottom line: Waymo/Cruise have demonstrated L4 robotaxis driving autonomously on difficult city streets. So far, Tesla has just given us a L2 driver assist. So right now, as it stands, Waymo/Cruise with lidar have achieved FSD and Tesla without lidar has not achieved FSD yet. I am not saying that Tesla will never do FSD. But that is the state of things right now.

I don't see Waymo in Atlanta. Oh yeah, only in one section of one city.

Elon has stated that they aren't going to play the game of autonomy only within this small area. It's only going to be LAX to NYC.

You can sit here and argue all you want. But until a car can actually drive where I want it to, it's all marketing speak. It's like all of these Electric vehicles that all of the manufacturers are going to have on the road in the next year. Until they start rolling off the production lines in quantity, they really don't exist in my book. I have yet to see a Ford Electric Mustang appear in the wild.

 

[diplomat33]

But until a car can actually drive where I want it to, it's all marketing speak.

It's all fine that Tesla is working on FSD that you can drive anywhere but Tesla has not delivered that either though. In fact, Tesla has yet to deliver any FSD at all. Waymo and Cruise have real autonomous driving. Just because you can't use it where you live, does not make it less real.

 

[ewoodrick]

Tesla has not delivered that either though. In fact, Tesla has yet to deliver any FSD at all.

Hmmm, it seems to be doing as well as the Cadillac, you know, what this thread is about.

I am not, nor ever have said that any autonomous product exists. My only premise, similar to Elon's s that detailed maps and LIDAR aren't the best answer.
AFAIK, the reality isn't really with the image recognition in either platform, it's the processing of what to do with it that becomes the real processing. Tesla's solution is showing today that it does an awesome job of reading the roads as well as most of the objects on the roads. Current releases are showing stop signs and traffic lights (whoops, LIDAR doesn't see color).

ALL, I repeat ALL, of the systems have false positives and negatives.

LIDAR has the advantage that it can provide relatively accurate distances to objects. But Tesla is also showing that vision systems can come pretty close. And when it gets down to the brass tacks, whether a car is 20m or 22m away isn't a big deal. But in the end, both systems draw a picture that is then fed into a "vision" system. A BIG part of a vision system is actually reducing the information that goes into the guidance system.

Why, if you are so entrenched in LIDAR are you on a Tesla forum? Why not go to another manufacturers platform that is supporting LIDAR? Can you remind me which manufacturer supports any portion of LIDAR in their vehicles today? I believe that Cadillacs are only using map data created by LIDAR mapping, not LIDAR itself.

 

[diplomat33]

Hmmm, it seems to be doing as well as the Cadillac, you know, what this thread is about.

Well, Supercruise and NOA both have their pros and cons. Supercruise is better in some ways. NOA is better in other ways.

My only premise, similar to Elon's s that detailed maps and LIDAR aren't the best answer.

And I disagree with that statement. Every company working on FSD uses them and has better FSD than Tesla. That would suggest that they are the best answer.

ALL, I repeat ALL, of the systems have false positives and negatives.

True but the question is how many.

Why, if you are so entrenched in LIDAR are you on a Tesla forum? Why not go to another manufacturers platform that is supporting LIDAR? Can you remind me which manufacturer supports any portion of LIDAR in their vehicles today? I believe that Cadillacs are only using map data created by LIDAR mapping, not LIDAR itself.

I am on this forum because I love my Model 3 and Tesla makes the best electric cars period. Also, AP is great, even if it is not FSD yet. It should be possible to love Tesla cars and love AP while still recognizing that Tesla's approach to FSD is not perfect.

 

[timewasted]

Why, if you are so entrenched in LIDAR are you on a Tesla forum? Why not go to another manufacturers platform that is supporting LIDAR? Can you remind me which manufacturer supports any portion of LIDAR in their vehicles today? I believe that Cadillacs are only using map data created by LIDAR mapping, not LIDAR itself.

Oh my god, I never in a million years would’ve guessed a post like this would be directed at @diplomat33! What a crazy world we live in.

@diplomat33 On this day, we are brothers!

Also, there was a conversation earlier in the thread about how “well what if a new bridge gets built?!” To anyone who feels like making such an argument, do you really believe that anyone who uses HD maps uses them exclusively? Everyone has a vision system. Everyone who has an actual chance of producing a self driving vehicle is using literally every tool at their disposal. Why? Because no one has ever made a level 5 autonomous vehicle, so no one knows exactly what is needed. Step 1 is make it work. Step 2 is optimize it.

One of these decades LIDAR will probably not be needed. This is not that decade.

 

[diplomat33]

Oh my god, I never in a million years would’ve guessed a post like this would be directed at @diplomat33! What a crazy world we live in.

@diplomat33 On this day, we are brothers!

Also, there was a conversation earlier in the thread about how “well what if a new bridge gets built?!” To anyone who feels like making such an argument, do you really believe that anyone who uses HD maps uses them exclusively? Everyone has a vision system. Everyone who has an actual chance of producing a self driving vehicle is using literally every tool at their disposal. Why? Because no one has ever made a level 5 autonomous vehicle, so no one knows exactly what is needed. Step 1 is make it work. Step 2 is optimize it.

One of these decades LIDAR will probably not be needed. This is not that decade.

I know, right. I used to be called a rabid fanboy. Now, I am being called a hater. LOL.

Look, I am not rooting for Tesla's FSD to fail. But I think we need to be real. Lidar and HD maps are useful. Like you said, most companies doing FSD use ALL the tools in order to maximize their chances at achieving safe FSD. And the fact is that right now, the companies that have opted for a full sensor suite (camera, radar, lidar) have demonstrated working FSD and Tesla which has opted not to use a full sensor suite has not yet demonstrated working FSD. That's just a fact of where things currently stand right now. So I just think it is a bit silly to argue that somehow the companies who are using all the tools and have demonstrated working FSD are somehow wrong but the company that has opted not to use all the tools and has not demonstrated working FSD yet, somehow really has the best FSD solution. That just does not make sense to me.

 

[diplomat33]

Since this thread is about Cadillac's Supercruise let me try to get things back on topic.

AP/NOA has some advantages over Supercruise but it is also true that AP/NOA on highway has had some issues. Every Tesla owner can attest to experiencing issues from time to time while using AP/NOA on the highway like phantom braking from an overpass, ping ponging in the lane when the an exit lane widens etc... HD mapping is a good tool to help reduce these issues. If you want to argue that Tesla can overcome these issues with better computer vision, that's fine. AP/NOA has actually gotten better over time. But to dismiss HD maps as useless when they would help make AP/NOA on the highway more reliable and safer, does not make sense to me.

 

[hugh_jassol]

What does it even mean to “use a lidar map“? A map is a semi static representation of routes, and lidar is used to visualize transient objects such as vehicles, pedestrians, bicyclists, etc.

It means they pre-mapped the freeways with LIDAR. That is, they drove a car around like google maps does and mapped all of the freeways for use with supercruise:

So they hired a company called Ushr to drive every mile of interstate in the US and Canada, mapping it with a lidar sensor.

source

 

[YYZ-IAD]

You can sit here and argue all you want. But until a car can actually drive where I want it to, it's all marketing speak

It's not all marketing speak. I think it's called a combination of extensive R&D, being very safety conscious in testing, and rolling out in stages. The smart thing to do.

Wall Street experts who only care about potential financial success, are putting huge valuations on the Autonomous Vehicle segment - and these aren't sentimental, dreamy marketing people. Latest estimate through Cruise is this segment is worth $8 trillion with a T. Uber does 5 billion trips a year. Take the driver out, and this unleashes over $70 billion in value per year - just for Uber. These are not university play-things. Hard work is being done to make big $ -- plus big social benefits providing mobility to those that lack it (elderly, etc).