GM’s new Ultra Cruise: Hands-free driving on all paved roads in US/Canada

[Bladerskb]

You are (in)correct on all counts. That is why GM has such excellent Battery Management Systems, and why the Bolt has never had a recall, nor had they ever introduced the V8.6.4 and so many more. That is why Boeing produced such superb reusable rockets, narrow body state of the art safety and outstanding mature technology on the 787. If spending more resources on the problem could be useful GM would indeed be leading, Boeing would ahem more orders than Airbus and SpaceX would have failed, not to mention Tesla.

Thoughtful criticism is both responsible and informative. Everyone who's been watching autonomous driving development is acutely aware that there is no practical solution today, and there is no assurance there may be any soon. Nobody really knows when it will be solved.

Things we do know today: 1. sensor incompatibility is a giant problem; 2.Update lags render radar ineffective. 3.. Lidar has exceedingly high sensitivity to precise and permanent labeling, making it both expensive and complex to use without both highly precise and highly accurate environmental data. Both rdar and lidar have some excellent applications but general over the road use is not one of them.

We do not know the limitations of vision based systems nor if the recent advances in utility and affordability will continue to evolve enough to cope with difficult atmospheric problems such as rain, snow, smoke and other pollution. We do not yet know how these systems can be made impervious to major lighting changes,

We do know that vision systems now can use electron microscopes, other visual systems that manage accuracy within a few microns, and rapid improvement in cost effectiveness of visual measurements that benefit from numerous variants of AI-based decision making. These developments demonstrate that eventhe surface of such approaches is far from defined.

I do not represent myself and authoritative in any way. I do claim to have a well defined sense of the obvious. The obvious, in my opinion, is that success will only come from computational advances and sensor improvements. Thus, I would bet on Andrej Karpathy and others like them who understand how to describe their process in terms that common people think they understand. They call these things like "Deep Learning" and "Computer Vision" that help get the funding needed to make faster and faster labeling of more and more obscure event descriptions so that they can train their tools to recognize undesired events faster and more reliability while reducing the risk of making a mistake and not recognizing something correctly. In that ancient statistical terminology, type one and type two errors.

The people who sell lidar and radar is preferred solutions haven't really understood the problem they are trying to solve.

Hence, they can easily make airport shuttle run on tracks and be perfectly Fien with no human driver. That approach can and does land a space vehicle quite perfectly on a space station. They can map the ocean floor. They cannot drive a car, unless it is in a tunnel.

Dojo to the rescue!!

This is pure nonsense

 

[alexgr]

No it doesn’t you are spreading misinformation. A heated camera can’t see in a blizzard. A heated radar can see perfectly in a blizzard. If you can’t understand the difference you shouldn’t be in this discussion.

This is why airplanes use heated radar.

Any radar will have difficulty seeing in rain, snow, or snow blizzard. The automotive radar (77Ghz-ish frequencies) radiation is absorbed by the atmosphere (water, gases) really well. This is one reason why it is not used in communications, and we can safely use automotive radars without interference with anything else.

 

[PDX-Y]

No it doesn’t you are spreading misinformation. A heated camera can’t see in a blizzard. A heated radar can see perfectly in a blizzard. If you can’t understand the difference you shouldn’t be in this discussion.

This is why airplanes use heated radar.

A driver* can't see in a blizzard, though, so I'm having trouble following the argument here. Instrument flight and autoland is indeed a thing for aircraft, it's not for cars. The design goal here is to automate everything a driver can do.

This argument seems to be getting out of hand, guys. Chill.

* Comes with built-in heat for its vision system. Also built-in cleaning/lubrication, blinking and eye-rubbing. The optional "baseball cap" and "sunglasses" packages are very popular.

 

[alexgr]

Which is why Tesla put heaters on their cameras which you were just boasting about? It’s quite clear to anyone using their common logic, if camera is clear it can’t still see in harsh weather. If radar is clear it can see through harsh weather like it’s not even there. To make radar/camera clear you put heaters on it, plain and simple.



No they haven’t. I never said they didn’t have heaters. They have it on the front and forward facing cameras on the b-pillars.

But other robotaxi cars not only have heating on all sensors, but also water and or air cleaning solution.

Tesla is using the traditional wipers for forward windshield, it’s not an engineered cleaning solution specifically for the sensors.

Unlike others...

https://twitter.com/x/status/1219923529543639040

Here is Mobileye’s air and water cleaner for all cameras.

And ofcourse you compare Tesla’s claimed “level 5 full self driving car” to Fords lane keeping assist car rather than to their actual robotaxis cars which ofcourse you have no idea about.

If Mobileye had sensors integrated under the windshield, they would not need to do anything else but to use a car wipers, and regarding backward facing cameras, I had zero issues with them ever, so I call it over engineering.

Also, on my Tesla, wipers automatically clean the windshield (with the washer fluid) when the camera detects dirt on the windshield even on a perfectly sunny day.

 

[EVNow]

“GM relied on lidar-scanned high-definition maps for Super Cruise, but Ditman said it wasn’t practical to map all 2 million miles of road for Ultra Cruise. “We do rely on similar map data,” he said. “However, we have a larger number of sensors that also observe the roads so when we combine the map accuracy with what our sensors see of the road geometry and the road markings, we’re still able to accurately place ourselves and drive the right nominal path.”

This is the important thing. Basically, Cruise is saying they won't use HD Maps - but basically a multi-sensor suit and 2-D Maps ? Does anyone have more information about this ? This would make Cruise different from both Waymo and Tesla.

 

[Bladerskb]

Just leaving this here. Ford’s Algo AI sensor cleaning test. It’s funny how people think their Tesla will summon to come pick them up cross country yet it doesn’t have basic self cleaning solution. With a rear cam that is completely obfuscated in mild rain/snow.

 

[EVNow]

Ford’s Algo AI sensor cleaning test.

Where do I buy this. If you can't then its not relevant - you posts are always anti-Tesla and OT. You work for a competitor of Tesla but doesn't have the basic integrity to disclose that information.

 

[stcptl]

You are (in)correct on all counts. That is why GM has such excellent Battery Management Systems, and why the Bolt has never had a recall, nor had they ever introduced the V8.6.4 and so many more. That is why Boeing produced such superb reusable rockets, narrow body state of the art safety and outstanding mature technology on the 787. If spending more resources on the problem could be useful GM would indeed be leading, Boeing would ahem more orders than Airbus and SpaceX would have failed, not to mention Tesla.

Thoughtful criticism is both responsible and informative. Everyone who's been watching autonomous driving development is acutely aware that there is no practical solution today, and there is no assurance there may be any soon. Nobody really knows when it will be solved.

Things we do know today: 1. sensor incompatibility is a giant problem; 2.Update lags render radar ineffective. 3.. Lidar has exceedingly high sensitivity to precise and permanent labeling, making it both expensive and complex to use without both highly precise and highly accurate environmental data. Both rdar and lidar have some excellent applications but general over the road use is not one of them.

We do not know the limitations of vision based systems nor if the recent advances in utility and affordability will continue to evolve enough to cope with difficult atmospheric problems such as rain, snow, smoke and other pollution. We do not yet know how these systems can be made impervious to major lighting changes,

We do know that vision systems now can use electron microscopes, other visual systems that manage accuracy within a few microns, and rapid improvement in cost effectiveness of visual measurements that benefit from numerous variants of AI-based decision making. These developments demonstrate that eventhe surface of such approaches is far from defined.

I do not represent myself and authoritative in any way. I do claim to have a well defined sense of the obvious. The obvious, in my opinion, is that success will only come from computational advances and sensor improvements. Thus, I would bet on Andrej Karpathy and others like them who understand how to describe their process in terms that common people think they understand. They call these things like "Deep Learning" and "Computer Vision" that help get the funding needed to make faster and faster labeling of more and more obscure event descriptions so that they can train their tools to recognize undesired events faster and more reliability while reducing the risk of making a mistake and not recognizing something correctly. In that ancient statistical terminology, type one and type two errors.

The people who sell lidar and radar is preferred solutions haven't really understood the problem they are trying to solve.

Hence, they can easily make airport shuttle run on tracks and be perfectly Fien with no human driver. That approach can and does land a space vehicle quite perfectly on a space station. They can map the ocean floor. They cannot drive a car, unless it is in a tunnel.

Dojo to the rescue!!

No one really knows who/which approach is MORE LIKELY to get there first? Really?

Again, a non-techie but it’s possible vision alone may solve FSD soon but it’s also possible it might take a very long time to improve vision enough to solve FSD. We don’t know yet (which is why it was wrong and possibly illegal for Elon to have SOLD FSD or L5 capability, his other accomplishments not withstanding). In case vision alone doesn’t do it soon, it’s possible, in conjunction with radar and/or LiDAR, vision could suffice. There may be those edge cases where vision is inadequate but radar/LiDAR augment it enough to make it workable.

So who has the riskier approach? Elon has made a huge gamble that he did not need to. I hope it doesn’t end up hurting progress on autonomy, EVs, tech etc Make it work in the lab, and then worry about scaling. Everyone else is managing to put in radars/lidars, Tesla could’ve too. If there is anything that will lead to Tesla getting there first, it’s likely their large customer fleet generating data.

 

[stcptl]

Where do I buy this. If you can't then its not relevant - you posts are always anti-Tesla and OT. You work for a competitor of Tesla but doesn't have the basic integrity to disclose that information.

I don’t think there is a need for personal attacks. People have differing opinions on the subject.

Maybe they are developing it for ridesharing and not for customers to purchase? In which case, their progress would not be publicly visible unlike Tesla’s.

 

[EVNow]

I don’t think there is a need for personal attacks. People have differing opinions on the subject.

LOL. This poster has no personal integrity - hate posts anti-Tesla all the time and is employed by a Tesla competitor. This is not a personal attack - a basic online etiquette.

When I worked for Microsoft, for 15 years never posted anything attacking any of Microsoft’s various competitors in any social media.

 

[PDX-Y]

Again, a non-techie but it’s possible vision alone may solve FSD soon but it’s also possible it might take a very long time to improve vision enough to solve FSD. We don’t know yet

Folks... we pretty much know. Just read this very subforum. Existing FSD with existing sensors is working. What bugs remain are essentially exclusively recognition and path planning issues. Detection of obstacles a-la LIDAR or Radar is, as far as we can tell within the sample size here (25k cars are running beta by my extrapolation), a solved problem.

The fight was had. Vision won. It's time to start yelling about other stuff.

 

[stcptl]

Folks... we pretty much know. Just read this very subforum. Existing FSD with existing sensors is working. What bugs remain are essentially exclusively recognition and path planning issues. Detection of obstacles a-la LIDAR or Radar is, as far as we can tell within the sample size here (25k cars are running beta by my extrapolation), a solved problem.

The fight was had. Vision won. It's time to start yelling about other stuff.

I hope you are right. I sure wouldn’t mind riding around in self-driving Teslas

 

[JHCCAZ]

Its hilarious when Tesla fans who wholly believe that ML can solve the motherload of AI problem...

Just a friendly note, the word is "motherlode" which comes from mining, the principal vein or lode of ore in the vicinity. A great word.

Yes I know that the other spelling is already in use, but this is primarily due to an understandable confusion of the original word, and its recent application to a Nintendo game. The latter now means a probably unrecoverable loss of the real word, but I want to put up at least some minor resistance on the way down...

 

[EVNow]

Its hilarious when Tesla fans who wholly believe that ML can solve the motherload of AI problem which is L5 self driving yet believe that mapping can't be solved/automated by ML.
(Hint its already been solved)

You should tell that to Waymo and others. After all Waymo still has people manually labeling exactly where a car should drive on every road and even alternate routes on parking lots. Tesla has 500 labellers and i’m sure Cruise and others have too.

When are you going to “grow up” from college level understanding of software to real world applications ?

 

[EVNow]

Mobileye already has ~virtually all of US mapped.

Link ? Have they mapped my neighborhood ?

Even 2D maps aren’t correct. I don’t exactly live in the woods - just 10 miles from world’s largest software company.

 

[jbcarioca]

Again, a non-techie but it’s possible vision alone may solve FSD soon but it’s also possible it might take a very long time to improve vision enough to solve FSD.

Were you to have a more technical background you might understand the importance of different time in response (radar being slow) and able to function without precise and accurate known environment (Lidar cannot do that, by design).

Many fundamental errors in major product design happen because of inadequate technical knowledge.
Understanding that does help avoid extrapolating from personal experience rather than known facts.
Many huge catastrophic failures have come from incorrect assumptions by even seemingly competent engineers.

As I have said, I am not an expert. I have helped diagnose many failures. nearly all were made by seemingly competent people who were superficial.

Hint: define vision
Hint #2: Human eyes are not examples of excellent vision

 

[stcptl]

Were you to have a more technical background you might understand the importance of different time in response (radar being slow) and able to function without precise and accurate known environment (Lidar cannot do that, by design).

Many fundamental errors in major product design happen because of inadequate technical knowledge.
Understanding that does help avoid extrapolating from personal experience rather than known facts.
Many huge catastrophic failures have come from incorrect assumptions by even seemingly competent engineers.

As I have said, I am not an expert. I have helped diagnose many failures. nearly all were made by seemingly competent people who were superficial.

Hint: define vision
Hint #2: Human eyes are not examples of excellent vision

I know, just that at this stage I trust Waymo et Al more than Elon

 

[EVNow]

I know, just that at this stage I trust Waymo et Al more than Elon

By your logic ( someone Berry being more beholden to stockholders who for apparently weird reason more worried about exactly when Cruise will be extended to city driving than EPS etc) - Tesla should be having battery issues. Not GM (or Nissan earlier).

In something like this (brand new technology that has not been solved) - you have to trust no-one. You won't be disappointed.

This from 2015 : (before someone says oh, they have robotaxi pilot, see what the implication of highlighted text is.)

Google's Autonomous Car Ready in Five Years?

Google believes only completely self-driving cars can fully address safety concerns.

www.pcmag.com

Google says it is well on its way to launching self-driving cars within five years. Because if they don't, Google director Chris Urmson's son will have to navigate the roads himself—and we can't have that.​

​

During this week's TED conference in Vancouver, Urmson told attendees that his eldest son, currently 11, is set to take his driver's test in a scant four-and-a-half years.​

​

"My team are committed to making sure that doesn't happen," he joked.​

 

[stcptl]

Time will tell!

 

[EVNow]

Time will tell!

Exactly. But the history on autonomous driving says everyone is wrong, so don't trust anyone.

Coming back on topic - what I really want to know is - is GM taking legal responsibility for city driving ?!?

That would indeed be a massive difference over Tesla and a "game changer".

This is why I hate these stupid "journalists" who act like stenographers. Absolute no relevant questions - just repeating marketing press releases dumbly. Would really like to have a AMA with Cruise/GM guys on this.