Google's Driverless Cars Legally Approved

[Juma]

In a manner of speaking. Radars can "see" a car in front of you even in dense fog/snow.
But we all heard a story of car's AI reacting to steam/fog.

Yes, driverless car is probably better than human in fog, but it's still much worse during nice sunny day. Human driver can look at pedestrian or other driver and determine what he/she is going to do next, computers can't do that well yet.

Describe a scenario where this matters.

Drunk pedestrian from Sweden.

And the scenario is?

 

[TV and credit cards]

In a manner of speaking. Radars can "see" a car in front of you even in dense fog/snow.

Latest thing I read was they were working with LIDAR now and snow reflection is still an issue.

 

[barbos]

In a manner of speaking. Radars can "see" a car in front of you even in dense fog/snow.

Latest thing I read was they were working with LIDAR now and snow reflection is still an issue.

bilby mentioned radars, they would not have such a problem.

- - - Updated - - -

In a manner of speaking. Radars can "see" a car in front of you even in dense fog/snow.
But we all heard a story of car's AI reacting to steam/fog.

Yes, driverless car is probably better than human in fog, but it's still much worse during nice sunny day. Human driver can look at pedestrian or other driver and determine what he/she is going to do next, computers can't do that well yet.

Describe a scenario where this matters.

Drunk pedestrian from Sweden.

And the scenario is?

Drunk pedestrian from Sweden.

 

[Juma]

Latest thing I read was they were working with LIDAR now and snow reflection is still an issue.

bilby mentioned radars, they would not have such a problem.

- - - Updated - - -

In a manner of speaking. Radars can "see" a car in front of you even in dense fog/snow.
But we all heard a story of car's AI reacting to steam/fog.

Yes, driverless car is probably better than human in fog, but it's still much worse during nice sunny day. Human driver can look at pedestrian or other driver and determine what he/she is going to do next, computers can't do that well yet.

Describe a scenario where this matters.

Drunk pedestrian from Sweden.

And the scenario is?

Drunk pedestrian from Sweden.

Details please. Present a detaiked scenario where the google car would do worse than the average driver.

 

[bigfield]

Drunk pedestrian from Sweden.

A scenario:

A drunk Swede is stumbling home along a footpath adjacent to a busy road where the speed limit is 80 kph. This pedestrian suddenly steers themselves off the footpath and onto the road, right at the moment when a car is approaching.

Human driver: Having taken no notice of a pedestrian walking along the footpath, the driver is taken by surprise by the pedestrian suddenly stumbling onto the road and slams on the brakes. However, the car still has too much momentum and hits the pedestrian at high speed.

Robot driver: Having tracked the pedestrian automatically since it was detected, the driver continuously updates the projected path of the pedestrian based on their latest trajectory. As soon as the pedestrian steers off of the footpath, the driver projects the collision and starts braking immediately. The car either stops short of the pedestrian, or hits the pedestrian at considerably lower speed than the human driver.


Let's see you provide a convincing argument for a scenario where the human driver outperforms the robot.

 

[barbos]

Drunk pedestrian from Sweden.

A scenario:

A drunk Swede is stumbling home along a footpath adjacent to a busy road where the speed limit is 80 kph. This pedestrian suddenly steers themselves off the footpath and onto the road, right at the moment when a car is approaching.

Human driver: Having taken no notice of a pedestrian walking along the footpath, the driver is taken by surprise by the pedestrian suddenly stumbling onto the road and slams on the brakes. However, the car still has too much momentum and hits the pedestrian at high speed.

Robot driver: Having tracked the pedestrian automatically since it was detected, the driver continuously updates the projected path of the pedestrian based on their latest trajectory. As soon as the pedestrian steers off of the footpath, the driver projects the collision and starts braking immediately. The car either stops short of the pedestrian, or hits the pedestrian at considerably lower speed than the human driver.


Let's see you provide a convincing argument for a scenario where the human driver outperforms the robot.

Human driver will detect drunk swede long before he decides to project his path into road way and will reduce the speed.
google cars do have troubles predicting pedestrians and other drivers.

 

[Jimmy Higgins]

A scenario:

A drunk Swede is stumbling home along a footpath adjacent to a busy road where the speed limit is 80 kph. This pedestrian suddenly steers themselves off the footpath and onto the road, right at the moment when a car is approaching.

Human driver: Having taken no notice of a pedestrian walking along the footpath, the driver is taken by surprise by the pedestrian suddenly stumbling onto the road and slams on the brakes. However, the car still has too much momentum and hits the pedestrian at high speed.

Robot driver: Having tracked the pedestrian automatically since it was detected, the driver continuously updates the projected path of the pedestrian based on their latest trajectory. As soon as the pedestrian steers off of the footpath, the driver projects the collision and starts braking immediately. The car either stops short of the pedestrian, or hits the pedestrian at considerably lower speed than the human driver.


Let's see you provide a convincing argument for a scenario where the human driver outperforms the robot.

Human driver will detect drunk swede long before he decides to project his path into road way and will reduce the speed.
google cars do have troubles predicting pedestrians and other drivers.

Yup. And we know that progress is impossible.

 

[barbos]

Human driver will detect drunk swede long before he decides to project his path into road way and will reduce the speed.
google cars do have troubles predicting pedestrians and other drivers.

Yup. And we know that progress is impossible.

No, I don't know that.

 

[bigfield]

A scenario:

A drunk Swede is stumbling home along a footpath adjacent to a busy road where the speed limit is 80 kph. This pedestrian suddenly steers themselves off the footpath and onto the road, right at the moment when a car is approaching.

Human driver: Having taken no notice of a pedestrian walking along the footpath, the driver is taken by surprise by the pedestrian suddenly stumbling onto the road and slams on the brakes. However, the car still has too much momentum and hits the pedestrian at high speed.

Robot driver: Having tracked the pedestrian automatically since it was detected, the driver continuously updates the projected path of the pedestrian based on their latest trajectory. As soon as the pedestrian steers off of the footpath, the driver projects the collision and starts braking immediately. The car either stops short of the pedestrian, or hits the pedestrian at considerably lower speed than the human driver.


Let's see you provide a convincing argument for a scenario where the human driver outperforms the robot.

Human driver will detect drunk swede long before he decides to project his path into road way and will reduce the speed.
google cars do have troubles predicting pedestrians and other drivers.

A human driver can only do that if there is adequate visibility, and more importantly, the driver is paying adequate attention to pedestrian traffic. Human drivers are notoriously inattentive, even when it comes to other vehicles.

Why do you say that Google's car has trouble predicting pedestrian movement?

 

[barbos]

Human driver will detect drunk swede long before he decides to project his path into road way and will reduce the speed.
google cars do have troubles predicting pedestrians and other drivers.

A human driver can only do that if there is adequate visibility, and more importantly, the driver is paying adequate attention to pedestrian traffic. Human drivers are notoriously inattentive, even when it comes to other vehicles.

Some yes, but most are pretty attentive, and AI currently sucks at it. AI can identify a human, and that's about it.
Google street view sometimes identifies a picture of human in a billboard as an actual human

Why do you say that Google's car has trouble predicting pedestrian movement?

Because they do.

 

[Juma]

A human driver can only do that if there is adequate visibility, and more importantly, the driver is paying adequate attention to pedestrian traffic. Human drivers are notoriously inattentive, even when it comes to other vehicles.

Some yes, but most are pretty attentive, and AI currently sucks at it. AI can identify a human, and that's about it.
Google street view sometimes identifies a picture of human in a billboard as an actual human

Why do you say that Google's car has trouble predicting pedestrian movement?

Because they do.

(In sweden we have a good saying for these nonsensical answers: "goddag, yxskaft" "hello, axe shaft" stemming from the story about a deaf axe shaft salesman that assumes that the first thing anyone asks him is "what do you sell".)

 

[bigfield]

A human driver can only do that if there is adequate visibility, and more importantly, the driver is paying adequate attention to pedestrian traffic. Human drivers are notoriously inattentive, even when it comes to other vehicles.

Some yes, but most are pretty attentive, and AI currently sucks at it. AI can identify a human, and that's about it.
Google street view sometimes identifies a picture of human in a billboard as an actual human

Why do you say that Google's car has trouble predicting pedestrian movement?

Because they do.

How do you know that? I would like to see a source.

 

[Malintent]

I can out drive it. I can avoid intentional threats. Can it? If you wanted to rear-end it, you could. Try to hit me and you'll miss.

No, I won't miss. There's this thing called Inertia and Momentum... Science always wins over belief.

 

[barbos]

Some yes, but most are pretty attentive, and AI currently sucks at it. AI can identify a human, and that's about it.
Google street view sometimes identifies a picture of human in a billboard as an actual human

Why do you say that Google's car has trouble predicting pedestrian movement?

Because they do.

How do you know that? I would like to see a source.

How do I know what? You imply that AI is already good, so you should give the source for that.
google car still relies on GPS and precise maps to know where it is, forget about reading intention of people on the road, they are just moving boxes to avoid for them.

 

[Tom Sawyer]

How do I know what? You imply that AI is already good, so you should give the source for that.
google car still relies on GPS and precise maps to know where it is, forget about reading intention of people on the road, they are just moving boxes to avoid for them.

Well, what problems do you feel their AI has in avoiding moving boxes? They can likely react quicker to changes in movement than that change in movement can alter the boxes' positions, so they don't need to read intentions, they just need to adjust to the new positions.

While it's true that the current AI couldn't pass a basic driver's test and there's a long way to go before they can really take over from humans, people moving about on the road seems like the most trivial of problems for the programmers to overcome.

 

[fromderinside]

How do I know what? You imply that AI is already good, so you should give the source for that.
google car still relies on GPS and precise maps to know where it is, forget about reading intention of people on the road, they are just moving boxes to avoid for them.

Well, what problems do you feel their AI has in avoiding moving boxes? They can likely react quicker to changes in movement than that change in movement can alter the boxes' positions, so they don't need to read intentions, they just need to adjust to the new positions.

Besides adding a bit of a doppler capacity should make their precision relative to other bodies much better than just GPS.

 

[Jimmy Higgins]

Here is an interesting cross technology problem. Google Car driving on highway, wants to pass old lady, but old lady has left her blinker on. How does the car handle that?

Some yes, but most are pretty attentive, and AI currently sucks at it. AI can identify a human, and that's about it.
Google street view sometimes identifies a picture of human in a billboard as an actual human

Why do you say that Google's car has trouble predicting pedestrian movement?

Because they do.

How do you know that? I would like to see a source.

How do I know what? You imply that AI is already good, so you should give the source for that.

We have reached an impasse boys. Send in the dogs!

google car still relies on GPS and precise maps to know where it is, forget about reading intention of people on the road, they are just moving boxes to avoid for them.

This is a good point. Google probably forgot to think of things like anticipating driver intent. Probably lucky they were able to build a car and not accidentally a robot hippo.

If I were designing an AI, I would model the other cars as wireline outlines moving along a vector. Funny, because that is my (and probably any one else that is paying attention) basis for driving, ie taking the vector of their travel.

One thing to keep in mind is that with GPS and maps, the car likely knows why certain vehicles are doing certain things, turning onto the road, merging on a ramp, crossroad, picking up hooker. With these sorts of knowns, you can establish baselines for what should and shouldn't be happening out there. When the computer picks up on an anomally, it can crunch some numbers and determine what the heck is happening and then adapt the vehicles movement.

 

[Malintent]

I don't think the Goggle car needs any turn indicator information... So it is not going to be stuck behind the old lady with her blinker permanently on. Turn indicators are for humans... it prepares you for what would otherwise be an unexpected obstacle in front of you all of a sudden... with Google's AI, it has no problem reacting to someone performing a lane change or slowing to exit... the turn signal is not needed... it simply detects the path the other driver is taking when they take it. The AI doesn't need a 'get ready to slow down' indicator... it just does.

 

[Jarhyn]

Y'all arguing over whether a Google car can identify a person, when we have toys that don't just identify persons, but individual body parts, and postures...

Seriously, how do y'all not remember this? It's kind of embarrassing.

Re: kinnect.

 

[Jayjay]

Y'all arguing over whether a Google car can identify a person, when we have toys that don't just identify persons, but individual body parts, and postures...

Seriously, how do y'all not remember this? It's kind of embarrassing.

Re: kinnect.

Only works indoors and up to certain distances. Of coruse there are other technologies available and image recognition is improving in strides, I just don't think it's quite there yet in terms of reliability for dealing with actual traffic.