I love Parallax's 360 feedback servo, but I want better resolution... the way to do this is a reduction gearbox. ServoCity has some nice ones... but really expensive.
I have to believe there is a market for such devices... 1 supplier?
I removed the pcb from standard servos and soldered longer wires to the ones inside. I attached the pot wiper to an ADC, the motor to an L293d and then I controlled the position with another pot. I controlled the speed with PWM.
I have also done the same thing by tying the shafts of larger geared DC motors to potentiometers. The only drawback was that in order to run multiple motors from one cog I had to write assembly code.
I did it because I wanted larger motors for a future robotic arm project. I liked it because in my tests there was less jitter in the movement.
I love Parallax's 360 feedback servo, but I want better resolution ...
What kind of resolution do you need? The encoder output is good to 12 bits (4096 divisions per revolution). The working resolution, of course, depends upon the Spin or C driver software and could be much coarser than that. But fine resolution is just a matter of different software, not gearing.
Parallax's feedback 360 servo is just about perfect for my application. I don't want to reinvent it, just dress it up a bit.
@Phil,
What I am trying to do is to build a gimbal that rotates a pair of PropCams around the center of each sensor. Think of it as a head with two eyes.... a very large head:)
.088 degrees sounds perfect... but that is an ideal number, which no doubt will be way off in my hands. I am planning for a certain amount of slop:) So, I was looking to reduce by 4... giving me plenty of room to wallow in my own mud.
You know how they are always saying..."our depth camera see's depth the same way that the human brain does"?
It isn't true...
For example the next generation RealSense cameras considers 64 possible offset positions ... and they have fixed optics.
The brain actually only considers about 15 different offsets, but uses convergence to pick where in space the sweet spot is... Around the convergence point objects can show plus and minus offsets, depending upon whether they are closer or farther than the convergence point. One thing a PropCam can't do is dynamically vary the focus. The brain uses active focus to enhance stereognosis.
I actually can ignore small errors, but if I am going to do it... I would like to do it right.
(if the price is right and it isn't too much of a hassle:)
Do you want to rotate each individual eye, or the pair as a unit? If the former, you lose parallax the further from center you turn, and your differential distance distortion increases. If the latter, the amount of parallax is invariant with the turning angle and there's no differential distance distortion.
I wrote this last night... and then forgot to post it:)
Ideally each individual eye should have 3 axis rotation around the center of each PropCam sensor ... and then both should be mounted on what would be similar to a normal pan and tilt mechanism. BUT I'm thinking this might be a nightmare to calibrate.
To simplify things, initially the two eyes will move together except around the y axis ... to allow "orthogonal" convergence... not exactly correct, but close enough to make the point.
converge independently... all other movement together. I think I have figured out... just need to find the right parts.
I don't like it, but it should work:)
Hmm, that surprises me, since the eyes, when they're shifting from far to near vision, behave as if they have gearing between the two. At least that's the mechanism I had in mind, which would require but one servo, not two.
Hmm, that surprises me, since the eyes, when they're shifting from far to near vision, behave as if they have gearing between the two. At least that's the mechanism I had in mind, which would require but one servo, not two.
-Phil
Not entirely right, just applies when the object is between both eyes. For something left or right from you center of vision the gearing with one servo would not work.
Now thinking of Marty Feldman or even chameleons converging independently is used in nature.
The thing is, if you want the image sizes between the two views to be the same, it's better to converge the "eyes" looking forward and turn the head. Obviously, human vision is able to compensate for the distance differences from just rotating the eyes; but computing image correlation in a computer when the binocular image features of interest are the same size is much easier than if they're not.
OK... I'll name it Feldman and throw in chameleon vision.
Just so we are using the same terminology: For camera movements relative to each other(two cameras), the motions are called vergences. When the movement of the cameras exactly aligns the center of view of each camera on a single point in space, it is called convergence. When the lens adapts to the distance of a point in space by active focusing, it is called accommodation. Fixation is the act of bringing a point in space onto to the center of view.
Chameleons are neat because they can diverge their fixation to give a wide field of view... or converge the eyes to allow stereo vision. Why they don't get dizzy? If we did this we would have such intolerable double vision, the brain would just ignore one eye.
@Phil,
I think we are saying the same thing... Think of it as stereo PropCams mounted on a pan and tilt... where the cameras are either parallel or converging toward a point.
Feldman optics has a nice ring to it. Just leave out the Marty...
If you do the main positioning thru fixation via pan and tilt, then @Phils approach with one servo for convergence will work. But for looking at moving objects the single pan/tilt might not be fast enough while combined with moving the convergence point might have a faster movement/tracking.
Basically you would need three Pan and Tilt gimbals, one for the whole head and one for each PropCam.
Does the P2 still have the functions to do coordinate transformations? This is part of the whole GEORZEL thing I still don't understand, right?
The chameleon version could track two objects independent!
Basically you would need three Pan and Tilt gimbals, one for the whole head and one for each PropCam.
Does the P2 still have the functions to do coordinate transformations? This is part of the whole GEORZEL thing I still don't understand, right?
The chameleon version could track two objects independent!
cool stuff,
Mike
That's right, ideally you need pan/tilt/rotation about z for both PropCams and then a pan/tilt/rotate for the whole kabaab. On the frst pass, I intend to cheat: just the y axis for the Propcams. And then a pan and tilt for the whole assembly... 4 degrees of freedom rather that 9.
The P2 should have no problems with coordinate transforms. I haven't done it yet, but I'm not worried. In my day, I was an excellent surgeon, but actually building stuff was never my strong suit. We'll see:)
Comments
I have also done the same thing by tying the shafts of larger geared DC motors to potentiometers. The only drawback was that in order to run multiple motors from one cog I had to write assembly code.
I did it because I wanted larger motors for a future robotic arm project. I liked it because in my tests there was less jitter in the movement.
-Phil
Thanks,
Parallax's feedback 360 servo is just about perfect for my application. I don't want to reinvent it, just dress it up a bit.
@Phil,
What I am trying to do is to build a gimbal that rotates a pair of PropCams around the center of each sensor. Think of it as a head with two eyes.... a very large head:)
.088 degrees sounds perfect... but that is an ideal number, which no doubt will be way off in my hands. I am planning for a certain amount of slop:) So, I was looking to reduce by 4... giving me plenty of room to wallow in my own mud.
You know how they are always saying..."our depth camera see's depth the same way that the human brain does"?
It isn't true...
For example the next generation RealSense cameras considers 64 possible offset positions ... and they have fixed optics.
The brain actually only considers about 15 different offsets, but uses convergence to pick where in space the sweet spot is... Around the convergence point objects can show plus and minus offsets, depending upon whether they are closer or farther than the convergence point. One thing a PropCam can't do is dynamically vary the focus. The brain uses active focus to enhance stereognosis.
I actually can ignore small errors, but if I am going to do it... I would like to do it right.
(if the price is right and it isn't too much of a hassle:)
-Phil
Ideally each individual eye should have 3 axis rotation around the center of each PropCam sensor ... and then both should be mounted on what would be similar to a normal pan and tilt mechanism. BUT I'm thinking this might be a nightmare to calibrate.
To simplify things, initially the two eyes will move together except around the y axis ... to allow "orthogonal" convergence... not exactly correct, but close enough to make the point.
-Phil
I don't like it, but it should work:)
Thanks
-Phil
Not entirely right, just applies when the object is between both eyes. For something left or right from you center of vision the gearing with one servo would not work.
Now thinking of Marty Feldman or even chameleons converging independently is used in nature.
Enjoy!
Mike
-Phil
just teasing, I did not say that this is a easy to solve software problem to use different angles, but nature did manage it. Somehow.
Mike
OK... I'll name it Feldman and throw in chameleon vision.
Just so we are using the same terminology: For camera movements relative to each other(two cameras), the motions are called vergences. When the movement of the cameras exactly aligns the center of view of each camera on a single point in space, it is called convergence. When the lens adapts to the distance of a point in space by active focusing, it is called accommodation. Fixation is the act of bringing a point in space onto to the center of view.
Chameleons are neat because they can diverge their fixation to give a wide field of view... or converge the eyes to allow stereo vision. Why they don't get dizzy? If we did this we would have such intolerable double vision, the brain would just ignore one eye.
@Phil,
I think we are saying the same thing... Think of it as stereo PropCams mounted on a pan and tilt... where the cameras are either parallel or converging toward a point.
If you do the main positioning thru fixation via pan and tilt, then @Phils approach with one servo for convergence will work. But for looking at moving objects the single pan/tilt might not be fast enough while combined with moving the convergence point might have a faster movement/tracking.
Basically you would need three Pan and Tilt gimbals, one for the whole head and one for each PropCam.
Does the P2 still have the functions to do coordinate transformations? This is part of the whole GEORZEL thing I still don't understand, right?
The chameleon version could track two objects independent!
cool stuff,
Mike
"Damn those Eyes" ... "Too Late"
That's right, ideally you need pan/tilt/rotation about z for both PropCams and then a pan/tilt/rotate for the whole kabaab. On the frst pass, I intend to cheat: just the y axis for the Propcams. And then a pan and tilt for the whole assembly... 4 degrees of freedom rather that 9.
The P2 should have no problems with coordinate transforms. I haven't done it yet, but I'm not worried. In my day, I was an excellent surgeon, but actually building stuff was never my strong suit. We'll see:)
Wish I could just buy the metal pinions which have internal servo spines. I can take it from there.
They are absolutely lovely... but just too pricey. I haven't found another source at any price. Quite a monopoly.
ServoCity also has something they call ServoBlocks... about $24. Very useful, but again no gear reduction.
I think I'm going to have to live with a resolution of about .088 degrees.
I found some nice servo turntables by Sainsmart and Lynxmotion... for less than $20, which pretty much solves the mechanical issues for me.
Still mulling it over... just can't believe I can't find what I want for about $3:)
Who are those guys?