Self driving cars - how do other makers besides Google do it?

[fuzzybabybunny]

So self driving cars don't actually do it entirely by sophisticated cameras and sensors and all that. They need a pre-mapped area to drive in.

http://mobile.nytimes.com/2013/12/1...global-domination-dont-ask-why-ask-where.html

Right now Google has about 25 experimental self-driving cars on public roads in California and Nevada. So far they have driven more than 600,000 miles without being involved in a serious accident. The self-driving algorithms do not work because there has been some breakthrough in artificial intelligence; they run on maps. Every road that Google’s robo-cars drive on was first surveyed by a human-driven pilot car outfitted with sensors accurate enough to measure the thickness of the painted lines in the middle of the road. Every detail of the road has been mapped beforehand. According to Peter Norvig, Google’s head of research, it’s a hard problem for computer vision and artificial intelligence to pick a traffic light out of a scene and determine if it is red, yellow or green. But it is trivially easy to recognize the color of a traffic light that you already know is there.

In effect, the robot car is not driving through the real world so much as it is moving through, in Borges’s words, “a map of the Empire, whose size was that of the Empire, and which coincided point for point with it.” When the real world is transformed into a data set, it starts to take on some of the aspects of the virtual.

Sergey Brin, Google’s co-founder, has promised to release self-driving technology within four years, and Google’s maps will then be a standard feature in its robot cars. Not to be outdone, Elon Musk has promised that Tesla Motors will deliver a self-driving car in three years. It’s too early to know whether Tesla will use O.S.M.’s maps — but the indications are that it will not use Google’s.

 

[Kaido]

There's a lot involved. For starters, if you get in a self-driving car, it simply needs to know where to go - so at the very least, it needs some sort of GPS. An up-to-date map would help tremendously; maybe if they partner with Amazon for weekly area drone mapping while making deliveries, haha.

There's the system built into the car, but then there's always network communication: traffic light sensors & traffic cameras, for example. And car-to-car communication. I can see a system like airplanes have (transponders) so they can help avoid each other. I don't see it ever taking off completely unless the government mandated it and including some sort of passive RFID transponder tag into license plates or something, because you have drug dealers, undercover officers, military transport, etc. who wouldn't necessarily want to be tracked.

It's a huge undertaking. Anxious to see it take off!

 

[Meghan54]

Watched a Top Gear episode where the military was utilizing a self-driving truck. Sensors all over, no mapping. Darned thing did rather well.

I'd tend to think it'll end up being a combination of gps and sensor data. Got to have sensors, how else is the vehicle going to pick up peds, other cars, unexpected road blockages, etc.?

 

[Midwayman]

I remember seeing an autonomous car demo awhile back that required you to drive the route at least once. It would build a database and use that to drive the route in the future. Great for commuting.

Read the article the excerpt posted kinda misrepresents the point. They're talking about the legacy of building up google maps with street view and all the google maps cars. Basically the GPS data we all know and love. Not the fact that they require some sort of 3d map to work. GPS map, yes.

 

[sontakke]

Helicopter one Google car in Mumbai and command it to drive from one end of the city to another. It won't make hundred feet before giving up and calling the helicopter to come and pick it up.

 

[fuzzybabybunny]

I remember seeing an autonomous car demo awhile back that required you to drive the route at least once. It would build a database and use that to drive the route in the future. Great for commuting.

Read the article the excerpt posted kinda misrepresents the point. They're talking about the legacy of building up google maps with street view and all the google maps cars. Basically the GPS data we all know and love. Not the fact that they require some sort of 3d map to work. GPS map, yes.

The point of that excerpt is that Google does more than map the street itself. It marks the spatial location of stop signs. The traffic lights overhead. The width of the road lines. The road signs. And then it integrates that data into the self driving car so the sensors already know where to look for the red/green/yellow light. If it didn't have this data, the sensors would have to somehow identify a red/green/yellow light anywhere and everywhere and at any time around the car, which is a much more overwhelmingly difficult task.

The point is that Google already has this tremendously detailed spatial map of every street it has driven and that can be used to greatly assist the other sensors and the self driving car.

How could a regular manufacturer, without access to this extensively mapped GPS spatial data, do as well?

 

[fuzzybabybunny]

Helicopter one Google car in Mumbai and command it to drive from one end of the city to another. It won't make hundred feet before giving up and calling the helicopter to come and pick it up.

Of course. Read the article or at least the OP.

 

[Sheep221]

There's a lot involved. For starters, if you get in a self-driving car, it simply needs to know where to go - so at the very least, it needs some sort of GPS. An up-to-date map would help tremendously; maybe if they partner with Amazon for weekly area drone mapping while making deliveries, haha.

There's the system built into the car, but then there's always network communication: traffic light sensors & traffic cameras, for example. And car-to-car communication. I can see a system like airplanes have (transponders) so they can help avoid each other. I don't see it ever taking off completely unless the government mandated it and including some sort of passive RFID transponder tag into license plates or something, because you have drug dealers, undercover officers, military transport, etc. who wouldn't necessarily want to be tracked.

It's a huge undertaking. Anxious to see it take off!

You can't use airplane transponder in road traffic, the cars are just too close to each other, so it would be constantly warning you pointlessly.

 

[Zorba]

The point of that excerpt is that Google does more than map the street itself. It marks the spatial location of stop signs. The traffic lights overhead. The width of the road lines. The road signs. And then it integrates that data into the self driving car so the sensors already know where to look for the red/green/yellow light. If it didn't have this data, the sensors would have to somehow identify a red/green/yellow light anywhere and everywhere and at any time around the car, which is a much more overwhelmingly difficult task.

The point is that Google already has this tremendously detailed spatial map of every street it has driven and that can be used to greatly assist the other sensors and the self driving car.

How could a regular manufacturer, without access to this extensively mapped GPS spatial data, do as well?

IMHO, if this is what it takes for a self driving car to work, it is nothing more than a gimmick. In the real world thing change, move, brake, etc all the time. Until a car can safely drive on any road in any condition with only a route programmed, I don't think they are ready for prime time.

This reminds me of a project I did in college. I was supposed to program a remote control car to scan around itself and make its way through a maze. Well this was the last project of my senior year, so I blew it off, as the deadline approached I realized I didn't know how to control the servo motor on the sensor and controlling the turn rate of the car was harder than I had expected. So finally, I just set everything up so it would work with the one maze design and hard coded everything.

 

[fuzzybabybunny]

Also the point is that we seem to be far from an actual self-enclosed self-driving system that operates in a vacuum with no external communication from other things.

The Google car operates well only because it is tied to GPS satellites miles above the earth and is receiving pre-mapped spatial data of permanent objects in its location to help it look for turnoffs and hanging traffic lights, for example.

I guess you could download the spatial data and access it locally, but it still needs GPS I think.

It is far from being a human, which really are self-enclosed decision making machines (and with all the inherent flaws of being so, I might add).

 

[Midwayman]

The point of that excerpt is that Google does more than map the street itself. It marks the spatial location of stop signs. The traffic lights overhead. The width of the road lines. The road signs. And then it integrates that data into the self driving car so the sensors already know where to look for the red/green/yellow light. If it didn't have this data, the sensors would have to somehow identify a red/green/yellow light anywhere and everywhere and at any time around the car, which is a much more overwhelmingly difficult task.

The point is that Google already has this tremendously detailed spatial map of every street it has driven and that can be used to greatly assist the other sensors and the self driving car.

How could a regular manufacturer, without access to this extensively mapped GPS spatial data, do as well?

You're reading a lot more into the article than is there. It says that the mapping cars were capable of detecting line width. Yes, they're camera. Doesn't mean it needs that info. Yes it's easier to detect traffic lights if we know where they are but we can make assumptions on where to look at road intersections, and it doesn't say its using 3d maps to locate them. The article author obviously is pushing an agenda here.

 

[Sheep221]

Also the point is that we seem to be far from an actual self-enclosed self-driving system that operates in a vacuum with no external communication from other things.

The Google car operates well only because it is tied to GPS satellites miles above the earth and is receiving pre-mapped spatial data of permanent objects in its location to help it look for turnoffs and hanging traffic lights, for example.

I guess you could download the spatial data and access it locally, but it still needs GPS I think.

It is far from being a human, which really are self-enclosed decision making machines (and with all the inherent flaws of being so, I might add).

It seems that world is almost completely covered by GPS maps, and with modern GPS deployment, where you have GPS in your phone, tablet, car etc I don't think that driving based on GPS data is so limited. You can go with that car readily everywhere so what's the problem?
Not to mention that even when you drive and have your GPS showing you way, you still get warnings about traffic problems, speed limits, signs, sometimes also intersections etc, if you can, why an autonomous cars shouldn't have this possibility?

 

[vshah]

I worked on darpa grand challenge for CMU back in 2005/6, we used pure computer vision/lidar along with gps and gyros. Of course those trucks were driving through the desert so it was OK to experiment since no lives were at risk.

 

[sontakke]

Look at how traffic moves in some third world cities. It is like utter and complete (but controlled) chaos. It will be many years before there is enough computing power in a self-driven vehicle to tackle the task.

Self-driving cars will work if the road only has self-driving cars. If there are pedestrians, hawkers, roadside stalls, bikes, auto-rickshaws, buses, cows, elephants, camels sharing the same road, you still need a human to do the driving.

 

[Sheep221]

Look at how traffic moves in some third world cities. It is like utter and complete (but controlled) chaos. It will be many years before there is enough computing power in a self-driven vehicle to tackle the task.

Self-driving cars will work if the road only has self-driving cars. If there are pedestrians, hawkers, roadside stalls, bikes, auto-rickshaws, buses, cows, elephants, camels sharing the same road, you still need a human to do the driving.

That's not the point, in that sense you take bus or taxi, autonomous vehicles are there to transport people with inability to drive just like they would be driving(without help of other person).