GM's self-driving car fails to see pedestrians and detects 'phantom' bicycles, report...

[Polycount]

In brief: Uber and Tesla aren't the only companies in the self-driving car industry who are facing some technical problems. GM and its autonomous car division Cruise are dealing with similar issues - according to anonymous sources, Cruise's vehicles have failed to detect pedestrians and have detected "phantom" bicycles that don't exist, prompting the cars to slam on their brakes erratically.

Carmaker General Motors acquired self-driving car startup Cruise way back in 2016, and it's made good use of its purchase ever since. Much like other companies in the tech and vehicle industries (such as Waymo, Uber, and Cadillac), General Motors (GM) views self-driving cars as the way of the future - as such, it's put quite a few resources into researching the tech and developing its own autonomous systems.

Indeed, in January, GM said it wanted to deploy fully autonomous, manual control-free vehicles by 2019; an ambitious goal, to be sure, but it began to look much more likely following GM's recent investment partnership with Honda.

Unfortunately, GM's plans may have hit a slight roadblock now. According to sources who spoke to Reuters, the carmaker's self-driving cars are having a difficult time identifying objects in motion, including pedestrians. Furthermore, the anonymous individuals said GM's vehicles have been known to see "phantom" bicycles, prompting them to slam on the breaks "erratically."

Given these (and other) hurdles, sources believe it is "highly unlikely" that GM will be able to hit its 2019 targets. Of course, considering the issues the industry has faced as of late -- including a handful of fatal self-driving crashes -- GM will likely not have any major qualms about delaying their plans if it results in a safer product for consumers.

At any rate, GM probably doesn't have much to worry about from its competition at the moment. Virtually every company in the autonomous car industry is facing similar problems (with the exception, perhaps, of Waymo), and few have come up with surefire solutions.

With this in mind, as consumers, we can probably expect at least a couple more years of testing and experimentation before we begin seeing true Level 5 automation in mass-market vehicles.

Permalink to story.

https://www.techspot.com/news/77083-gm-self-driving-car-division-facing-technical-challenges.html

 

[DaveBG]

All other manufacturers are still "testing the waters" , have no idea what are they doing and are doing it wrong. Tesla is so far ahead that at one point it will become hilarious to watch them struggle to understand the world they ended up in.

 

[Polycount]

All other manufacturers are still "testing the waters" , have no idea what are they doing and are doing it wrong. Tesla is so far ahead that at one point it will become hilarious to watch them struggle to understand the world they ended up in.

I'm still hoping for full self-driving tech from Tesla sometime in 2019. Probably a pipe dream, though.

 

[wiyosaya]

All other manufacturers are still "testing the waters" , have no idea what are they doing and are doing it wrong. Tesla is so far ahead that at one point it will become hilarious to watch them struggle to understand the world they ended up in.

I'm still hoping for full self-driving tech from Tesla sometime in 2019. Probably a pipe dream, though.

I have to agree. Even if they occur infrequently, Tesla has some serious issues with their autopilot which is not even supposed to be an autonomous driving tech.

 

[ShagnWagn]

These things need to be forced onto the public - whether the public fears for their safety or not! Even with needless and mericless deaths, it should be government-mandated all cars are self-driving by 2022.

 

[MilwaukeeMike]

All other manufacturers are still "testing the waters" , have no idea what are they doing and are doing it wrong. Tesla is so far ahead that at one point it will become hilarious to watch them struggle to understand the world they ended up in.

I don't think they are ahead - I think they called their driver-assist technology 'Auto-pilot' and it sounds like it's autonomous. Hard to predict anything about Tesla though since Musk is now banished to building the hyper-loop and work on visiting Mars for the next year.
We just had the story within the last week or so that Cadillac's driver assist systems beat Tesla's and Cadillac is GM.

 

[OutlawCecil]

Important fact to remember: Auto-pilot cars only need to slightly reduce accidents for them to be worth it. Fact: people won't be satisfied unless they are perfect even when human drivers are far from perfect.

 

[IAMTHESTIG]

Please fix the grammatical error, its brakes, not breaks.

Anyway, I don't believe self-driving cars will ever work until they put more sensors in vehicles and perfect computer based driving through many years of work.

It is good to see this companies sensor array is more capable than some of the competition (Tesla, Uber) but I feel they still need more. Doesn't appear they are using any thermal sensors which could be a big help in some situations.

Once you have all the possible sensor data then your software has to know how to interpret it and react accordingly, keeping false positives to a absolute minimum. No one is going to want to ride in a vehicle that is random braking because it thinks it saw a bicycle. And no bicycle rider wants to get hit by a self driving car because it didn't see them. It all comes down to having sufficient data and good software.

We'll be lucky if someone figures that out by 2030, honestly.

 

[Cycloid Torus]

Most likely it is merely bad maths. Gee-I'm-a-tree.

 

[Evernessince]

Important fact to remember: Auto-pilot cars only need to slightly reduce accidents for them to be worth it. Fact: people won't be satisfied unless they are perfect even when human drivers are far from perfect.

That's not even counting that these kinds of cars will let people with disabilities get around or the many convenience factors they will bring. Getting out of the grocery store? Press a button on your phone and have it pull up to the front, ready the instant you step out the door.

No need for a driver's license, hours of additional free time instead of driving, and you never have to worry about getting a ticket or ending up in jail due to driving again.

You'd have to be crazy to not see the benefits of self driving cars for everyone.

 

[DaveBG]

I don't think they are ahead - I think they called their driver-assist technology 'Auto-pilot' and it sounds like it's autonomous. Hard to predict anything about Tesla though since Musk is now banished to building the hyper-loop and work on visiting Mars for the next year.
We just had the story within the last week or so that Cadillac's driver assist systems beat Tesla's and Cadillac is GM.

They have momentum that is not visible. That is why some reports suggest they are still not N1 too. But they have the best fleet and the most collected road data + the best hardware installed in even the oldest models. Its a matter of OTA update and will be done in an Instant

 

[Sean T]

You would think that a group of manufacturers would get together, combine their resources and agree to share the tech amongst themselves instead of each individual company trying to build their own solution from scratch.

 

[gamerk2]

You would think that a group of manufacturers would get together, combine their resources and agree to share the tech amongst themselves instead of each individual company trying to build their own solution from scratch.

Except the ones who invested out of the gate would lose their advantage, which they are using to try and prop up car sales.

 

[gamerk2]

I see nothing but "false detection" problems ahead. I work in the world of RF, and car companies are starting to run into the same problems we've run into for decades now: How to distinguish what is real, and what isn't. Seeing something that isn't there leads to unnecessary slowdown, but the opposite case is much more worrying.

At the end of the day, I doubt car companies will invest in the necessary signal processing tools necessary to make their tech reliable.

 

[Theinsanegamer]

All other manufacturers are still "testing the waters" , have no idea what are they doing and are doing it wrong. Tesla is so far ahead that at one point it will become hilarious to watch them struggle to understand the world they ended up in.

I'm still hoping for full self-driving tech from Tesla sometime in 2019. Probably a pipe dream, though.

Given tesla no longer lets you buy the full driving package anymore, its not even a pipe dream.

 

[IAMTHESTIG]

That's not even counting that these kinds of cars will let people with disabilities get around or the many convenience factors they will bring. Getting out of the grocery store? Press a button on your phone and have it pull up to the front, ready the instant you step out the door.

No need for a driver's license, hours of additional free time instead of driving, and you never have to worry about getting a ticket or ending up in jail due to driving again.

You'd have to be crazy to not see the benefits of self driving cars for everyone.

Some very good points. I totally agree there are a lot of advantages to this tech, as long as it works well; but it will never be perfect and that is what I want. I'm not trying to hate here, I want this tech too but I just think companies using all the available technogies at their disposal, which can save lives. And when you think of yourself, or a family member or other loved one getting killed because a self-driving car didn't see them you will start to wonder what could have been done to prevent it. Humans aren't perfect, and frankly our driving schools in the US are appallingly basic and teach almost nothing about actual intelligent, defensive driving but at least humans can be held accountable if they hit/kill someone. A computer, cannot. THAT is the reason I seem so anti-self-driving tech.

 

[IAMTHESTIG]

You would think that a group of manufacturers would get together, combine their resources and agree to share the tech amongst themselves instead of each individual company trying to build their own solution from scratch.

Perhaps, but we all know how difficult it is to control the direction once you have so many people involved in its development. A small company with a handful of employees can work much more closely together than a giant monster of companies combined efforts. Furthermore we've seen some amazing innovations when there is competition in the marketplace. A lot of things wouldn't exist if companies weren't constantly trying to one-up the other and theoretically make a better product.

 

[IAMTHESTIG]

I see nothing but "false detection" problems ahead. I work in the world of RF, and car companies are starting to run into the same problems we've run into for decades now: How to distinguish what is real, and what isn't. Seeing something that isn't there leads to unnecessary slowdown, but the opposite case is much more worrying.

At the end of the day, I doubt car companies will invest in the necessary signal processing tools necessary to make their tech reliable.

Indeed... and RF can provide inconsistent results of the same thing in different environments. This is why there needs to be a combination of many different sensors including optical, IR illumination, and thermal. But it will all come down to software making decisions, which may be impossible to perfect. All this tech is very expensive for one vehicle so it just isn't practical to make it. So we are all stuck with basic use of sensors.

 

[Capaill]

Please fix the grammatical error, its brakes, not breaks.

Correct. Now find and fix 4 grammatical errors in your post

 

[gamerk2]

Indeed... and RF can provide inconsistent results of the same thing in different environments. This is why there needs to be a combination of many different sensors including optical, IR illumination, and thermal. But it will all come down to software making decisions, which may be impossible to perfect. All this tech is very expensive for one vehicle so it just isn't practical to make it. So we are all stuck with basic use of sensors.

And from what we can see, the SW isn't up to snuff. Fact is, image processing isn't anywhere near the point where we can say "That's a trash bag; run it over", and be absolutely sure we just didn't run down grandma. Throw that on top of all the radar/lidar processing issues, and I am very pessimistic this is anything more then a fad.

I fully expect that after one or two high profile accidents that regulations will clamp down due to public outcry, resulting in the tech getting dropped.

 

[IAMTHESTIG]

Correct. Now find and fix 4 grammatical errors in your post

Not until someone fixes the article. I'm just an ***** community commentator, not a professional news writer. ;-)

PS, I did spot three errors in my original post.