[WriterOfMinds]

Guess I haven't started a project thread for this one yet? Anyway, I put a bunch of work into the mechanical eyeball this year. It's been tougher to get it working perfectly than I expected (when is it not?), but I'm getting closer. Lots of photos and design discussion in the blog, so much I won't try to copy it all here: https://writerofminds.blogspot.com/2023/11/atronachs-eye-2023.html

Problems worked on this year and late last year include slack in the tendons, too much friction between ball and socket, limited range of motion and limit-of-motion sensing.

[MagnusWootton]

That's a ball joint actuator,  I never could make one of these...  I cant even think of it.
I just put three 1d joints in series, but thats not entirely the same thing.

Thanks for posting this,   I'll give it a better read later on cause I actually do with some help to make a ball joint!  Hardest joint to make!!!

[MagnusWootton]

I see it is tendon actuated.  with this I would say the range of freedom is down a bit possibly on first sight with knowing nothing much,   what is the angular range your getting out of this?

[WriterOfMinds]

I haven't measured it, but it's less than the full 180 degrees along each dimension. It's good enough for my purposes (I don't want the eye to roll over so far that the camera goes behind one of the eyelids and gets blocked).

[MagnusWootton]

90 would be useful to me.   more than that I count as very flexable.

[MagnusWootton]

If you had say 24 of these, like on a multi-eye spider robot head, and they are all trained on a different body part.  (like one on each eye, nose and corners of mouth,  then the shoulders, elbows, and knees) You'd have less occlusion/ line of sight problems up close to the robot cause they are all pointing different directions, you could tell people apart from the feature distances, plus you could get full mocap happening on the person for even more things for the robot to do, understand.

[WriterOfMinds]

Something fun I've thought about is to have eyes on little drone platforms, wirelessly connected to the main robot body. But that's a theoretical project for the far future, probably. First, I have to finish this eye!

[MagnusWootton]

that would give the robot a good advantage,  because occlusion disturbs robots more than people cause of being based apon strict methods, that could bring back the whole place fully 3d tracked, and things being in the way wouldn't affect it as much.

I actually might make one myself now,   this idea seems cool for computer vision, I think ill try doing it like an old mouse (sorta backwards tho) with little rollers.   See what happens.

[WriterOfMinds]

Update! I got some problems with the motion tracking worked out. Switched from MOG2 background subtraction to Farneback Optical Flow as my method of finding motion in the visual field. Accuracy seems improved and I have a new video of the eye following me back and forth across the room. https://writerofminds.blogspot.com/2024/10/atronachs-eye-2024.html