Self driving car project

[iacas]

On July 1, 2016 at 1:54 PM, Lihu said:

1) Sensors are primitive.

2) GPS not accurate enough.

3) Visual SLAM technologies are still in infancy.

4) Obstacle avoidance still uses heuristic processing, and might need to be enhanced with some cognitive ability to work properly under all conditions. Networking with applications like "Waze" are still in their infancy.

5) Roads are too unpredictable and might require some more cognitive skills to navigate properly without a persistent map of the region.

6) It will be difficult to figure out all the laws and insurance regulations within a shorter period of time.

7) I'm not convinced that people are willing to give up their driving privileges that soon.

I think the tech would be ready in 5-10 years… the problems will be:

• the laws will lag behind. The politics will lag way behind.
• the people will still want to drive. They won't "trust" the cars to drive themselves, and they won't like that the cars obey the speed limits.

[nevets88]

11 hours ago, iacas said:

I think the tech would be ready in 5-10 years… the problems will be:

• the laws will lag behind. The politics will lag way behind.
• the people will still want to drive. They won't "trust" the cars to drive themselves, and they won't like that the cars obey the speed limits.

I agree people still want to drive, but in congested areas, I think less so. Traffic jam? Bring on the robot overlords. NOW. I... HATE... DRIVING...

[boogielicious]

I think it will be huge for seniors who can't or shouldn't drive anymore. They will still have their independence and not have to rely of getting a ride.

[iacas]

4 hours ago, nevets88 said:

I agree people still want to drive, but in congested areas, I think less so. Traffic jam? Bring on the robot overlords. NOW. I... HATE... DRIVING...

I can't remember the last time I was in a traffic jam, but I don't live in NYC.

Maybe with self-driving cars we can get fluid dynamics that approach that of water. When lanes go from 2 to 1 for construction, cars double the speed to keep the flow rate constant.

That'd beat the pants (and underwear) off what happens now…

[nevets88]

23 minutes ago, iacas said:

I can't remember the last time I was in a traffic jam, but I don't live in NYC.

Maybe with self-driving cars we can get fluid dynamics that approach that of water. When lanes go from 2 to 1 for construction, cars double the speed to keep the flow rate constant.

That'd beat the pants (and underwear) off what happens now…

One of the examples where self driving is self evidently an improvement. 

Rubbernecking.

People slow down to a crawl just to look at a routine, banal, nothing to see here accident. Even when it's just a breakdown or minor fender bender, nothing spectacular -like a car dangling precipitously on a divider, half on one side, half on the other, like a see-saw - people slow down to look.

My environment strongly influences my opinion on this subject. I'm in a traffic jam almost every... f***ing... time... I drive in the area. I know what it's like to drive in less congested areas, it's awesome, but I'm past the point of no return. Drivers are super aggressive, unlike in the UK and other parts of the world, where civility still ensues on the road, relatively - even Canadian cities emit a bit of nastiness on the roads.

Like amber waves of grain, the American iconography of open top, wind in face, carless perfectly flat paved endless roads, radio blaring, trophy wife in the passenger seat is a thing of a very very distant past.

Just teleport me.

[saevel25]

On 7/1/2016 at 1:54 PM, Lihu said:

The current technology to drive is still very primitive. Most of it is based upon the Darpa grant challenge projects. The mapping and localization systems are not accurate enough yet. In another 3 years the technology might be adequate, but it will take a lot of work testing under all conditions. It will likely take more than just mapping and simple obstacle avoidance systems to drive safely. There might be a requirement for some cognitive reasoning especially if there are many manually drive cars around. So, technology-wise it could take yet another 30 years for slightly more cognitively intelligent robots.

BMW, Intel, and Mobileye collaborated effort have projected they will bring a fully automated vehicle to production in 2021. That would be 5 years away. 

[ZappyAd]

On 03/07/2016 at 0:29 PM, boogielicious said:

I think it will be huge for seniors who can't or shouldn't drive anymore. They will still have their independence and not have to rely of getting a ride.

I agree - and I could see as I get older I would likely be more interested in 'car as a service' rather than the hassle of owning a vehicle outright.

Also in rural areas for public transport you can have a fleet of smaller vehicles that run more 'on demand' rather than one big bus that only goes on a predetermined route and spends a lot of time empty or less than half full.  Plus you get more value out of your vehicles as they can be more easily run 24/7, and in theory they last longer with less maintenance and lower running costs as they will be driven more efficiently.  It might not be for everyone but the benefits for a large number of people could be significant.

 

[newtogolf]

10 hours ago, saevel25 said:

BMW, Intel, and Mobileye collaborated effort have projected they will bring a fully automated vehicle to production in 2021. That would be 5 years away. 

How will these vehicles perform in bad weather given my backup camera is sometimes blocked by snow and distorted during heavy rain.  Also what about roads where lane markings are barely visible or non-existent.  

[saevel25]

6 minutes ago, newtogolf said:

How will these vehicles perform in bad weather given my backup camera is sometimes blocked by snow and distorted during heavy rain. Also what about roads where lane markings are barely visible or non-existent. 

That is an issue with video detection in northern states. More so with snow rather than rain. I have no clue how the video cameras will actually perform under more extreme situations. I am not an engineer at BMW.  

Here is the technology Mobileye is selling. I am sure they are creating proprietary technology for BMW. 

http://www.mobileye.com/en-us/technology/

 

 

 

[newtogolf]

3 hours ago, saevel25 said:

That is an issue with video detection in northern states. More so with snow rather than rain. I have no clue how the video cameras will actually perform under more extreme situations. I am not an engineer at BMW.  

Here is the technology Mobileye is selling. I am sure they are creating proprietary technology for BMW. 

http://www.mobileye.com/en-us/technology/

I wasn't actually expecting you to have the answers just posing those questions as a bit of skepticism regarding their timeline.  

A good part of my career was spent in the imaging arena, mostly document imaging but some video analysis too.  I'm sure there is better technology today than I used but the algorithms to process images and video are still very CPU and memory intensive.  For automated driving applications where analyzing real time video data is critical I'd think they'd need redundant systems which will not only pose cost issues but cooling concerns as well.  

Then these is all the fun AI issues to consider, if braking isn't an option does the car run over small animals, deer, people or does it run itself off the road.   I don't know if five years is enough time to work out all the AI issues.  

[saevel25]

6 minutes ago, newtogolf said:

Then these is all the fun AI issues to consider, if braking isn't an option does the car run over small animals, deer, people or does it run itself off the road.   I don't know if five years is enough time to work out all the AI issues.  

That is going to be an issued that needs to be discussed out in terms what is acceptable programmed behavior. If the vehicle can safely maneuver out of a situation with out a collision I am sure that is being looked at. 

Certain situations are just not preventable. Specifically if a kid runs out in front of a vehicle and the time is so short that it is physically impossible to stop the vehicle in time. 

In the end if a vehicle is able to detect a person and brake sooner than a human. That speed reduction could in fact mean that a person might live from being struck by a vehicle. 

If a vehicle can reduce the speed from 30 mph to 20 mph by the time a human even starts braking then the chance of a human surviving the collision is improved from 55% to 95%. 

[newtogolf]

35 minutes ago, saevel25 said:

That is going to be an issued that needs to be discussed out in terms what is acceptable programmed behavior. If the vehicle can safely maneuver out of a situation with out a collision I am sure that is being looked at. 

Certain situations are just not preventable. Specifically if a kid runs out in front of a vehicle and the time is so short that it is physically impossible to stop the vehicle in time. 

In the end if a vehicle is able to detect a person and brake sooner than a human. That speed reduction could in fact mean that a person might live from being struck by a vehicle. 

If a vehicle can reduce the speed from 30 mph to 20 mph by the time a human even starts braking then the chance of a human surviving the collision is improved from 55% to 95%. 

Agreed, I'm very pro technology but I do have some concerns that AI could be potentially making decisions on who lives or dies.  A good driver that sees a deer or person running out in front of them can more quickly evaluate their available options than video analysis can.  

I'm looking forward to real testing, as this is a huge step for us.  

[nevets88]

While I think there are a lot of issues to be ironed out with self driving cars, I think we give human operators a little too much credit as to their infallibility. I see so much bad driving out there. Drivers running stop signs on purpose, sometimes two, three in a row. Texting, make-up, eating, looking at the gps, heated conversations, kids in the back making a fuss. While computers have some progress to make, so do humans, especially Americans in taking its driving rights for granted. It is really a privilege to be able to drive so freely as we do in the US. One thing computers don't have, it's emotions. They don't drive with their emotions.

[nevets88]

NTSB's preliminary report, with a photo of the crashed Tesla. It was going 74 mph.

http://www.ntsb.gov/investigations/AccidentReports/Pages/HWY16FH018-preliminary.aspx

 

[Lihu]

On 7/26/2016 at 1:02 PM, nevets88 said:

NTSB's preliminary report, with a photo of the crashed Tesla. It was going 74 mph.

http://www.ntsb.gov/investigations/AccidentReports/Pages/HWY16FH018-preliminary.aspx

 

Yeah, this was just dumb. . .there should have been a limiter since sensors are not capable to detecting things reliably 2-3 seconds out at those speeds. The sensor noise is far too high. Those idiots at Tesla are going to set back self driving cars a couple more decades if they don't smarten up. . .

[iacas]

7 hours ago, Lihu said:

Yeah, this was just dumb. . .there should have been a limiter since sensors are not capable to detecting things reliably 2-3 seconds out at those speeds. The sensor noise is far too high. Those idiots at Tesla are going to set back self driving cars a couple more decades if they don't smarten up. . .

I hardly think the guys at Tesla are "idiots."

The guy ignored all directions and was doing something like reading a book instead of paying attention. They're not fully self-driving cars, and you're not supposed to use them as he did.

[Lihu]

13 hours ago, iacas said:

I hardly think the guys at Tesla are "idiots."

The guy ignored all directions and was doing something like reading a book instead of paying attention. They're not fully self-driving cars, and you're not supposed to use them as he did.

Sure, operator error is the cause of >99% of any accident even those "found" to be due to manufacturing/design defects. 

Also, I'm not referring to the engineers at Tesla, but to whomever told them not to put an automatic cruising limiter at 55mph or less. I'm pretty sure the engineers would not recommend allowing the vehicle to automatically pilot the vehicle in excess of 45mph to 55mph with a live subject in or around it. Often times, engineers are told to do some pretty nebulous stuff in the name of "marketing". However, I also have to admit that a 45mph to 55mph or less limit for the automatic cruising feature might not sell that well, but that's probably about the limit of current technology. Likely even less depending upon the exact sensor technology they are using. . .

One of my ex-colleagues works there in a scientist role, I suppose I could ask him the details that he is allowed to divulge.

[saevel25]

5 minutes ago, Lihu said:

Sure, operator error is the cause of >99% of any accident even those "found" to be due to manufacturing/design defects. 

Where did you pull this stat from? 

6 minutes ago, Lihu said:

Also, I'm not referring to the engineers at Tesla, but to whomever told them not to put an automatic cruising limiter at 55mph or less.

http://www.theverge.com/2016/7/26/12287118/tesla-autopilot-crash-images-speed-limit-ntsb

There is a limiter, but it's only active residential roads and roads without a center divider.

So, to claim it wasn't put on is false. They have that feature. It's just not fully 100% on all driving conditions. 

If I had my choice, I would say that it should cap the speed at 5 over the speed limit or 70 mph which ever is lesser for any street.