I wonder if Google can recognize the single finger gesture and what it's algorithm says to do.....

They are building an experience base with the trips they are taking and continuing to refine their algorithms. I recently read that the Google car can recognize pedestrian hand gestures.

In the example you give, the car will have significant advantage over a human because the radar will be very precise as to distances and it will be less likely to overreact, being able to assess the situation on a millisecond by millisecond basis.

In situations where the algorithm has to track multiple moving cars, I think that a computer based solution with radar measurements will always be superior to a human. What I am concerned about is when something really weird happens, like a hang glider landing in front of you, bolders rolling down a hill, hazards in the road.

Oh they do.Their forums are full of people who vent on all kinds of issues with the game, from controls to AI behavior. The coolest thing is they are way outside the box on what can or cannot be done and they are coming up with some really nifty ways for AI on all levels to interact, predict and behave. It's not just human AI either, the bits of the game are testing platforms for this stuff, so the AI running the part where you go after a swarm of bad guys has to be able to counter your efforts and fight back realistically. They are no where near perfect, but the lessons they learn as people play the different sections are giving them mounds of data. They have a race section where you race against other players or AI's , the space itself is treated as an AI (using physics that was specifically created to make the bits and pieces work as if really in space, not just look like it). All that will lead to better and better decision algorithms which will port over to real world use I believe.

Star Citizen...need to look that up...

Just did. So you are suggesting that programming a game AI to fight against human controlled opponents is a viable step towards improving AI vehicles. I think that is a great idea, and you will get many hundreds of hours of testing by people who are just enjoying themselves playing the game.

I could not see if the gamers who were themselves programmers had a venue to suggest/write improvements to the AI.

Jan

Jan, you are absolutely right, an AI can be developed to be able to deal with unstable human reactions and interactions. I am sure military battle computers are much better at "prediciting" opponets possible options. I am a huge fan (and have a sizable fleet) of Star Citizen, and one reason is they actually understand the need for AI's in a persistant universe to be able to be able to function with regard to unstable humans around them and be believable. I am betting the logic they are building for SC will someday (and after being patented for them) be standard in both games and devices. If you have never seen SC got to Roberts Space Industires and check it out.

Due to this thread, both Wm and I have been looking up the stats on driverless cars. It is stated that not one of the accidents was the fault of the driverless car - but I do not have a complete description of each incident. It would be wonderful to see road tests where the type of action you describe was tested on a track: multiple (driverless) cars in a lane; the last one accelerates to pass and then cuts in.

It occurs to me that your misgivings are both reasonable and testable - there IS an answer to that question. In Star Trek and B5, humans control the actions of a battle - both flying and firing. It will probably be AI that does that in the real future...the great-grand-daughter of the driverless car.

Jan

While that is perfectly sound logic, my concern is that intentional actions may cause issues with logical decision making. Example: Dude screams by wanting to pass several cars, see oncoming truck, pulls in front of you (who's robotic car is exactly the correct distance behind the one ahead) and your robotic car slams on the brakes, in addition, the dude slams on his brakes because he closing the car ahead of him. Happens a lot in liberal Oregon, where some people seem to think they have special privileges to go where they need to at Warp 9.9. I do not know if I trust a program to adequately understand such non standard maneuvers and not do something rash in response.

The difference I see between a autodrive car and a human driving a car is twofold: attention and reflex time. It is a given that there will be manually controlled cars on the road, and that some of them will be driven by idiots. But if the car you are in is autocontrolled, it will have faster reflexes and never get tired or have an attention lapse.

So, if you hold 'idiot drivers' constant, then increasing the abilities of the other cars (via autodrive) should improve overall road conditions and decrease accidents.

Jan

I was thinking a monorail could have many spurs and junctions, with looping connections to get cars where they need to be based on usage. Would really be better than the infamous light rail they pander...Start with heavily traveled corridors and expand out just like roads. Volume build a reducing cost basis and once paid for should be inexpensive. Just need the cars and drives to be really well built for endurance.

You must believe that you can be alert to the other guy 360 degrees all the time. I actually think a computer will do a better job at that than I will.

As to the monorail, I would rather go where I want to go than where the monorail goes. I can walk out my door, get in my car and it will take me where I want to go.

At some point, it will be able to drop you off where you are going and go look for a parking place. Call your car on your cell phone and it will come and get you.

Indeed, I believe it would be nice, but I still have issues with the "other" guy. If all cars were on it, ok. But not yet. I myself believe that a monorail system, computer controlled with 2-4-6 person capsules that can be stored around the system, would be a more efficient system. Use recycled plastic to make the poles and rails. The Seattle one has always been a favorite of mine and seems so easy to design and work with,

If you are driving 62 miles to work each day, there must be times when you are tired or inattentive.

Imagine if you could spend those two plus hours each day reading while the car drove you. You might even get online and post to the Gulch.

My concern is that, even though they have sensors and all kinds of logic, humans manage to screw up anything. They are completely unpredictable and do things for the oddest reasons. I would say the first death toll will sound when they are more numerous and someone is texting or watching a movie and they get in an accident, and sue the hell out of the manufacturer.Their claim will be the car should have known the idiot on the left was going to swerve into the lane in front, cause they were texting too, and the car should have known. Or something like that. I drive 62 miles to work each way, for 3-4 days a wekk, and have at least one encounter with someone who "has to go faster" and passes on solid lines, blind curves or even around farm machinery. I guess we will have to see..

Four accidents under 10 miles an hour. Two of them when a human was in control -- which makes it not a self driving car. So, the self driving cars when they are driving themselves are as safe as when humans are driving them.

Humans driving cars are very dangerous, people get killed every day doing it.

The cars are already doing much better than people. For example, they know not to start immediately when the light turns green because one idiot could be running the light. They won't get tired, be inattentive, under the influence etc. They will always pay attention to everything.

This technology promises to make it safer, and will eventually not even need our attention -- the first generation will still require people behind the wheel.