Musings About AP 2.0

[SageBrush]

I'm more than willing to accept that AP2 can self-drive when other cars and the environment are acting predictably, but I am highly skeptical of an AP car acting smartly when the world is crazy. E.g,

This morning I took my usual drive home on a rural highway that has one lane in each direction. I was heading north, and a stream of cars were passing southward. I watched in my rear-view mirror as a truck came up way too fast behind me and then swerved into the opposing traffic lane to pass. It was immediately obvious that there was no where near enough time for the truck to pass me and return to our lane without colliding head-on with the traffic heading our way even though I was slowing my car down to give the truck more time. The driver in the opposing lane saved the day by also slowing down, and driving off the road onto his shoulder.

So not only did the smart driver figure out the best course, he realized that illegal driving was called for. That sounds pretty darn hard to program, let alone the infinite variations that can occur in real-time.

 

[SilverRose]

So not only did the smart driver figure out the best course, he realized that illegal driving was called for. That sounds pretty darn hard to program, let alone the infinite variations that can occur in real-time.

I was just having a conversation about AP just last night. One of my friends was pondering the idea of the next generation of drivers and if there would still be a use for kids to get their driving licence. And if there would ever be a time when we would not need driver's licence.

I think perhaps in time, the more AI is programmed into AP anything is possible but if we are going to be realistic about the autopilot feature it is right now just Auto Assist. I think that if it was to change there would still have to be some form of regulation to hold the owner responsible for the car, so if the car decided to turn into Christine and start killing people the owner would be held accountable.

I think in time the cars will be able to auto adjust and take illegal actions to lean towards the side of safety and I don't think that time is soon. People are slow to change in society and they are still struggling with electric cars.

 

[SageBrush]

AP safety is a non-linear dynamic. As it becomes more common it will get better in and of itself, but also true just having less crazy and unpredictable drivers will help.

So depending on your POV, this issue of safety is either a chicken and egg problem or a self fulfilling prophecy.

 

[SilverRose]

Do you feel that the AP will become more prevalent? If so do you feel that there will be a time when the average person won't actually know how to drive a stick or understand the breaking and steering?

 

[SageBrush]

People will always understand breaking

 

[Buddy]

Do you feel that the AP will become more prevalent? If so do you feel that there will be a time when the average person won't actually know how to drive a stick or understand the breaking and steering?

I wouldn't say the average person knows how to drive a stick even today. I don't see a time in this generations future that drivers are allowed to operate a motor vehicle without understanding how to brake or steer. I also don't see drivers training changing until every vehicle on the road is capable of autonomous driving....and THAT I don't see happening for a very, very long time.

 

[Carl]

Had a similar experience as the OP this afternoon. Approaching an intersection where I 'legally' had priority, I noticed a car approaching to cross my path. The car was going very slowly but the driver was on the phone, obviously quite enervated, and not ever looking in my direction, so I wisely let him pass. That was based on my understanding of where he was looking, and my appreciation of his apparent state of mind. My first thought afterwards was that (as long as there are humans driving cars) AP would really have to learn a lot about human nature.

(we could also, at a certain stage, simply forbid humans from driving cars on public roads, of course *)

* but given that AP does 0-60mph in around 20 seconds, I'm not sure what that would do to sales of the P100D...

 

[sxnahm]

While I have no doubt that developers of autonomous driving vehicles will be able to implement the "down the middle" driving cases, I wonder how they will handle the many edge cases vehicles encounter.

A couple weeks ago I needed to drive from SF to Berkeley during rush hour. I was using Waze to pick the route, and just as I was about to enter the freeway Waze pops up a notice that there is a stoppage on the freeway and reroutes me to enter it closer to the Bay Bridge.

Long story short, I spent the next several hours inching towards multiple entrances with no joy, until at last I was inline for the last SF entrance before the bridge. As I sat in a long line of cars on Folsom Street, I see an SF traffic control officer walking down the line of cars and speaking to the drivers. After a short chat the car makes a U-turn and takes off. His message: the entrance you are heading for is closed, enter on 5th or 8th (both of which I previously had tried to use). I follow his advice and finally got on the freeway, cross the bridge and arrived at my destination very late.

How would an autonomous vehicle deal with this? Would it just give up and go home at some point? How would a traffic control officer advise the vehicle that the route is not feasible? It's not hard to come up with "edge cases" that the autonomous software has to deal with, as this thread has demonstrated.

Regarding the officer interaction, I don't think Tesla's AP 2.0 has microphone input, so perhaps the officer can hold a sign up in front of the car that says "Go to freeway entrance at 8th Street" for the car to decode and act on. It's going to be a brave new world.

 

[sxnahm]

Regarding the officer interaction, I don't think Tesla's AP 2.0 has microphone input, so perhaps the officer can hold a sign up in front of the car that says "Go to freeway entrance at 8th Street" for the car to decode and act on. It's going to be a brave new world.

Actually, the first thing the officer did was ask me, "where are you going?" I told him to Berkeley and then he advised me to take another route. Should autonomous vehicles have a means to communicate with traffic officers regarding destination and intended route?

 

[ItsNotAboutTheMoney]

While I have no doubt that developers of autonomous driving vehicles will be able to implement the "down the middle" driving cases, I wonder how they will handle the many edge cases vehicles encounter.

A couple weeks ago I needed to drive from SF to Berkeley during rush hour. I was using Waze to pick the route, and just as I was about to enter the freeway Waze pops up a notice that there is a stoppage on the freeway and reroutes me to enter it closer to the Bay Bridge.

Long story short, I spent the next several hours inching towards multiple entrances with no joy, until at last I was inline for the last SF entrance before the bridge. As I sat in a long line of cars on Folsom Street, I see an SF traffic control officer walking down the line of cars and speaking to the drivers. After a short chat the car makes a U-turn and takes off. His message: the entrance you are heading for is closed, enter on 5th or 8th (both of which I previously had tried to use). I follow his advice and finally got on the freeway, cross the bridge and arrived at my destination very late.

How would an autonomous vehicle deal with this? Would it just give up and go home at some point? How would a traffic control officer advise the vehicle that the route is not feasible? It's not hard to come up with "edge cases" that the autonomous software has to deal with, as this thread has demonstrated.

Regarding the officer interaction, I don't think Tesla's AP 2.0 has microphone input, so perhaps the officer can hold a sign up in front of the car that says "Go to freeway entrance at 8th Street" for the car to decode and act on. It's going to be a brave new world.

That's actually a different problem. Priority 1 should be stopping people from heading that way using electronic communications that allow rapid dissemination of information. Then that's what the autonomous car would use. But even without that, an autonomous car has to have a navigation system and therefore you would have a way to tell it that you need a diversion. The big challenge would be if the autonomous car is _empty_ and heading somewhere. Then the car needs at least to have a number the cop can call to tell them about the problem so that they can push the closure to the car's navigation system. (Companies and agencies that interact with the police usually have identification mechanisms to make sure that it's genuine information.)

 

[ItsNotAboutTheMoney]

I'm more than willing to accept that AP2 can self-drive when other cars and the environment are acting predictably, but I am highly skeptical of an AP car acting smartly when the world is crazy. E.g,

This morning I took my usual drive home on a rural highway that has one lane in each direction. I was heading north, and a stream of cars were passing southward. I watched in my rear-view mirror as a truck came up way too fast behind me and then swerved into the opposing traffic lane to pass. It was immediately obvious that there was no where near enough time for the truck to pass me and return to our lane without colliding head-on with the traffic heading our way even though I was slowing my car down to give the truck more time. The driver in the opposing lane saved the day by also slowing down, and driving off the road onto his shoulder.

So not only did the smart driver figure out the best course, he realized that illegal driving was called for. That sounds pretty darn hard to program, let alone the infinite variations that can occur in real-time.

I don't think that it's that hard. Rule #1: avoid collisions. Rule #2: obey the law.

Your example rests on identifying the vehicles and doing some math. I think that the case described by another person where they avoided a potential collision with an inattentive driver might be more challenging since it might rest on anticipating driver behavior. Assuming plenty of processing power, if the car can identify all the vehicles around and their speeds, it should be able to calculate threats and the path out of danger. Besides, an autonomous car should be really good at playing chicken.

 

[Pdub2015]

Massive processing power, expanded sensor capabilities, instantly accessed database of millions of similar scenarios, direct communication with traffic advance warning communications... AP has many advantages over human wetware. We just need to be patient, as it's going to take some time...