zarthrag
Newbie level 6
Greetings everyone, I've browsed this forum a little, and I must say I'm very impressed! I can't believe I've gone this long without finding this place, a thumbs up to the moderators here! :wink:
Anyway, I wanted to start a discussion about methods for incorporating vision in robotics, ultimately for the purposes of AI. It would be quite the challenge for me to accomplish this, but I'll try anyway:
My friends and I have modified some R/C cars for uP control. The driving AI's we wrote *simulate* wonderfully! They can turn, pass, and even play a little rough. But there is a really big problem when it comes to sensory issues. To put it simply, I'm dealing with a blind Jeff Gordon here. I want this car to be able to navigate a dynamic track and not have any problem, even if the track has to have loud neon colors on the walls--that's ok--but It looks like I'm being forced towards optics here.
What's the cheapest way to incorporate video, should I just cut up some webcams? Will I have to dedicate an entire PIC processor to the cause. And once I get the video, how on earth could I begin decoding all of that data! A 320x200 image would contain a wealth of data that would be old before I decoded/understood it all!
I'm thinking about using CdS cells along the sides of the vechicle to detect brightly colored stripes (neon tape or walls). And maybe if they are movable, it can see what's down the road....But then again, a forward facing camera can do the trick if placed correctly.
:robot: - Zarthrag
Anyway, I wanted to start a discussion about methods for incorporating vision in robotics, ultimately for the purposes of AI. It would be quite the challenge for me to accomplish this, but I'll try anyway:
My friends and I have modified some R/C cars for uP control. The driving AI's we wrote *simulate* wonderfully! They can turn, pass, and even play a little rough. But there is a really big problem when it comes to sensory issues. To put it simply, I'm dealing with a blind Jeff Gordon here. I want this car to be able to navigate a dynamic track and not have any problem, even if the track has to have loud neon colors on the walls--that's ok--but It looks like I'm being forced towards optics here.
What's the cheapest way to incorporate video, should I just cut up some webcams? Will I have to dedicate an entire PIC processor to the cause. And once I get the video, how on earth could I begin decoding all of that data! A 320x200 image would contain a wealth of data that would be old before I decoded/understood it all!
I'm thinking about using CdS cells along the sides of the vechicle to detect brightly colored stripes (neon tape or walls). And maybe if they are movable, it can see what's down the road....But then again, a forward facing camera can do the trick if placed correctly.
:robot: - Zarthrag
