PDA

View Full Version : [Interesting] How would you create a terrain sensor?????



elios
09-28-2008, 11:03 PM
how would you create a terrain sensor that could tell wood from concrete and bark? what about water?

Would it be possible for the robot to create a 3d map inside its brain and track where all the surfaces are, so it automatically knows whats safe to approach and whats unsafe?

Would you use some sort of IR establishment where the quality of the ray determines the how reflective the surface is from other surfaces, using a IR resistor???

Does this make sense at all?

lnxfergy
09-29-2008, 09:31 AM
Ok, so nobody is responding to this one - I believe that is a confirmation of what I was thinking: such a sensor is very difficult. In fact, it will likely have to include several sensors, and even then, it will likely not work well.

You could get a rough idea using vision. Color segmentation could put you in a realm of materials (i.e. gray includes concrete, etc...). If you then looked at texture you could further refine it. Water however, is going to be tough to "see" - perhaps have a feeler with a water sensor (I'm guessing such a thing exists for those alarms that tell you when your basement is flooded... but I have no real idea). Or maybe use your camera, in concert with an arm - shove the arm into the surface, see how far it goes after the tip stops moving in the camera - and voila you have the depth of the water.

As to the map, that's fairly standard - look up what is known as an "occupancy grid" map. You could simply mark cells as DANGER or OK. However, your robot will need some very good feedback to know where it is in the map at all time (i.e. encoders and high traction tires, probably some good PID on the motor control).

A final note... water may not be your only danger, what if the grass is really high, or the shag rug is really thick - the robot may get stuck.

-Fergs

Enusi Malik
09-29-2008, 10:51 AM
I came across this topic a little while ago in "Scientific America Reports" (Spacial robots edition)
One of the techniques, was that the robot made a 3D map by using stereoscopic computer vision as it went down the hallway. CMU used a technique in the 1980's used a sensor combination to distill a large amount of of noisy sensor data to make maps of its surrondings, but the can only acheve 2D imaging. And the other goes down memorized route. Sorry for the brief description soooo I gotta a link for ya on the tugger :-) hope this helps a wee bit. I'll look up some info to see if I can get som details.
http://www.reuters.com/article/pressRelease/idUS230070+03-Apr-2008+MW20080403
peace

Adrenalynn
09-29-2008, 12:52 PM
Yeah, that's all ancient history. The going deal is LIDAR for 3D mapping these days - but not for terrain specifics like not falling in a puddle. Not too many puddles on Mars [unfortunately]. ;)

I've done a couple projects I've posted here. One that might have some promise for that would be a spectrometer - much cheaper and easier to build than the name suggests. ;)

You could also use a probe as Fergs suggests - as long as you're not worried about falling into distilled water...

I totally agree with him, and I thought the same thing when I first read this: No single sensor exists to do what you're looking to do. It has to be a sensor array.

4mem8
09-29-2008, 01:38 PM
This is cool stuff, and I am all ears as to the replies to this question. A difficult one as far as I can see. elios: I can see you are brewing something up, Got inspired by your visit eh!

Matt
09-29-2008, 01:57 PM
I'm thinking you have to use binocular vision for stuff like this. Such a system would have to be able to identify objects and textures by sight, just like most of nature. I'm always skeptical of trying to "cheat" it with other sensors, but obviously it makes sense to use such things until the software comes along that is smart enough to rely only on vision. That is a long way off in my humble opinion.

You could make a robot with an arm that can point and have it ask its human buddy. A human sensor if you will :P "Can I go there?" "How about there?" "What is that?" "What is this?" "Can if go here?" "How about over here?" LOL

lnxfergy
09-29-2008, 02:11 PM
Actually, the easiest solution is probably to make it so that nothing is off limits. Make the robot a cross between a car/boat/submarine and then give it a parachute for when it falls off a cliff. Bubble wrap the exterior, seal the insides up water tight, and paint it a bright color for human recovery in the case it does die.....

-Fergs

robot maker
09-29-2008, 02:33 PM
will most likely need a very long arm for that type of test,using a depth sounder will be better,like sonar design added to a camera,
new sensors are beening made every day it seems,like i read about a nose sensor that was made


Ok, so nobody is responding to this one - I believe that is a confirmation of what I was thinking: such a sensor is very difficult. In fact, it will likely have to include several sensors, and even then, it will likely not work well.

You could get a rough idea using vision. Color segmentation could put you in a realm of materials (i.e. gray includes concrete, etc...). If you then looked at texture you could further refine it. Water however, is going to be tough to "see" - perhaps have a feeler with a water sensor (I'm guessing such a thing exists for those alarms that tell you when your basement is flooded... but I have no real idea). Or maybe use your camera, in concert with an arm - shove the arm into the surface, see how far it goes after the tip stops moving in the camera - and voila you have the depth of the water.

As to the map, that's fairly standard - look up what is known as an "occupancy grid" map. You could simply mark cells as DANGER or OK. However, your robot will need some very good feedback to know where it is in the map at all time (i.e. encoders and high traction tires, probably some good PID on the motor control).

A final note... water may not be your only danger, what if the grass is really high, or the shag rug is really thick - the robot may get stuck.

-Fergs

science_geek
01-22-2009, 09:42 AM
how about a an ultrasonic range sensor that finds the distance to something, and then a laser range sensor to get the actual distance, if the distance of the ultrasonice range sensor is different from the laser, you can find the density of the object, just like a fish finder can see the bottom of a lake, but also can see the scummy bottom. as for water. you could scan the ground a certain distance away and then do a sweeping arc, if the difference between the laser and ultrasonic sensor is different by a certain amount, than there would be a puddle. you could mount the laser and ultrasonic sensor to a pan tilt mount and move fast so long as you avoid going faster than the echo