Well that was clever

cluelessPHP

Apparently predicted the accident before it happened
[video=youtube;nbAOSJ4t99U]https://www.youtube.com/watch?v=nbAOSJ4t99U[/video]

laserlight

I perceive it more as a warning that an accident is probable rather than an outright prediction: it looks like the accident was avoidable at the time of the alarm.

cluelessPHP

Perhaps, the most interesting thing about it I think is the amount of calculations that are most likely done each second, I like the idea of it

NogDog

I suspect by far the hardest part of auto-piloting cars is dealing with human drivers in other cars. Once all vehicles on the road are computer controlled and "talking" to each other, it has the potential, at least, to be much safer than just people driving. I read somewhere that one of the test programs (maybe Uber's?) had the result that their vehicles had to be programmed to be conservative and expect the worst from other vehicles, which made them have to hesitate and/or go slower much more than human drivers normally would.

Weedpacket

One of the tenets of defensive driving is "Everyone else on the road is an idiot", after all.

NogDog

I think it was George Carlin who said something like, "Have you ever noticed that everyone who drives slower than you is an idiot, and everyone who drives faster than you is a maniac?"

sneakyimp

I think NogDog and Carlin said it best.

The speed of the Tesla processor isn't really the story in this case. Think about it: even your cell phone (or a $35 raspberry PI) can process a few hundred million instructions per second or more these days. Chips are fast. What is difficult is the identification and tracking of multiple objects. I'm guessing there's some kind of spatially mapped array of pixels (a camera? a laser?) that has to be repeatedly scanned to identify and distinguish moving objects (which is hard) and determine which ones are coming close and which are not (hard) and furthermore identify a threat situation. It's a pretty gnarly programming problem.

cluelessPHP

sneakyimp;11059429 wrote:
I think NogDog and Carlin said it best.

The speed of the Tesla processor isn't really the story in this case. Think about it: even your cell phone (or a $35 raspberry PI) can process a few hundred million instructions per second or more these days. Chips are fast. What is difficult is the identification and tracking of multiple objects. I'm guessing there's some kind of spatially mapped array of pixels (a camera? a laser?) that has to be repeatedly scanned to identify and distinguish moving objects (which is hard) and determine which ones are coming close and which are not (hard) and furthermore identify a threat situation. It's a pretty gnarly programming problem.

Certainly seems great, I wonder how it's done, might look into it in about ten years after I get the basics right :p

sneakyimp

This looks helpful:
http://cs.stanford.edu/people/teichman/papers/arso2011.pdf

cluelessPHP

sneakyimp;11059445 wrote:
This looks helpful:
http://cs.stanford.edu/people/teichman/papers/arso2011.pdf

Looks interesting, will sit and read it, although those boys at Stanford make me want to go out and start apologising to trees :rolleyes:

cluelessPHP

Well sort of sounds similar to an idea I had awhile ago, was talking to another student about it basically it went along the lines of

If you could find a way to scan redshift and blueshift's of a surrounding area whilst also detecting gravity waves you might be able to create a 3d image outside of them frame, sort of like you see in films, but as I was told by the lecturer who overhead me I'm not google so I left it as a stupid idea

sneakyimp

I expect something similar to redshift and blueshift could be acquired with a laser scanner, but such information would not be easily extracted from a simple camera image. Detecting distinct objects in a video feed is probably a matter of determining which pixels are roughly consistent from frame to frame -- the trick being how to track an object as it jumps a few pixels to left or right when you hit a bump. I'd bet it involves some statistical approach.