/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

Roadmap: file restoration script within a few days, Final Solution alpha in a couple weeks.

Sorry for not being around for so long, will start getting back to it soon.

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB


(used to delete files and postings)

Robot Vision General Robowaifu Technician 09/11/2019 (Wed) 01:13:09 No.97
Cameras, Lenses, Actuators, Control Systems

Unless you want to deck out you're waifubot in dark glasses and a white cane, learning about vision systems is a good idea. Please post resources here.



Edited last time by Chobitsu on 09/11/2019 (Wed) 01:14:45.
>Unless you want to deck out you're waifubot in dark glasses and a white cane
But, OP, what if the model for the waifubot is supposed to be blind? #triggered
Open file (9.23 KB 480x360 0.jpg)
The inmoov project managed to use the playstation move(?) to give the robot vision, the script is freely available for it. They also use purchasable cameras for the eyes.


That's cool anon thanks. I had planned on using the open sauce version of kinect (Willow Garage) and a couple of hi quality 1080p webcams.
Open file (22.08 KB 480x360 0(1).jpg)
I just purchased a couple of the JeVois cameras. I plan to try using them on a little moebot. What's neat about them, is the have a dedicated 4-core CPU running Linux and any other vision software like OpenCV right in the camera. This pretty much entirely offloads the vision processing computation from the robowaifu's other onboard processors. Thanks to the anon in other thread for first posting the camera for us.

Blindfolded girls a cute.

On topic: a new Google vision kit
Thanks anon, I'll check it out come New Year. If I feel it doesn't feed their botnet, I'll recommend it. Either way, I'll post back here and let you know what I think about it in roughly a month or so.

>ed. surely anon will deliver...
facial recognition convo:
Anon linked a nice OpenCV training source:

OpenCV tutorial site

JeVois camera set up:
I just found this;
I'm thinking of using an acrylic light guide / light splitter that reads a single image from a TFT display and duplicates it for two small eyeballs. The eyeballs can be physical, so this will save some logic and wonkiness and I can just focus on making a nice retina pattern; the light guide can be a bunch of optical fibers. Pic is supposed to be a Lego piece, wish there was a better illustration.
Sounds interesting FluffyDev, can't say I really 'get it' specifically just yet, but I am familiar with optical fibers.
Upon VERY SERIOUS consideration, one-eyed girls may work for our purpose, either temporarily one-eyed or even permanently, since the space behind the other eye can be used for additional electronics and actuators (much needed space especially for a smaller 80cm bot such as mine). A tube guide consisting of either solid plastic tubing or bundled optical fibers and a mirror system will make it possible to have both a real camera vision system as well as fancy pupil/retina graphics. Plus with a physically moving eye, the camera will be able to pan also. (Yes it will be a huge eye mechanism, hence the SERIOUS CONSIDERATION of making the waifu one-eyed on purpose). But she will actually have vision which you can display on the debug computer!
>not just having a leela robowaifu
plebeian pls
Eye patch is beauty
I'm not really a fan of this eye mechanism since it doesn't support a camera but it could be redesigned to support one. The eyes look pretty good and the process could be adapted to creating anime doll eyes too.

>How to Make Realistic Eyes Using 3D Printing for Animatronic Eye Mechanisms

>How to Build a Simple 3D Printed Arduino Animatronic Eye Mechanism
Yes, I think I remembered that one from before. I like the face the parts are mostly 3D-printable with a cheap printer. Can probably use a resin-based UV printer for the more precision parts like the eyeballs/lids themselves. I don't see any reason small cameras couldn't be fitted inside the interior globes of the eyeballs anon. Even the Jevois camaeras (that have actual tiny fans on them b/c Linux+OpenCV coprocessor right on board) should work with the half-open design of the eyeball.
>I like the fact*
Open file (2.07 MB 320x228 790.gif)
reminds me of the Robot from Lexx
Here's one with two cameras and low amount of space: https://youtu.be/DY4as7Lc9KY My own thinking is, that ideally we should be able to take eyeballs out from the front. The sockets should be expandable four maintenance. If taken out, the servos should be replaceable as well. But then, I'd like my waifu to be waterproof (shower) and she should have tears. So the servos need to be separated with a layer of silicone somehow, maybe they move a magnet and the eyeballs have some metal inside?
>>4335 >that ideally we should be able to take eyeballs out from the front. The sockets should be expandable four maintenance. That's a great idea Anon. It will allow for easy upgrading of the cameras, and as the control mechanisms will be moving thousands of times a day (even if only for small distance) affords maintenance access. I propose that the entire assembly be able to slide forward on a tray to afford easy access from most angles.
>>4337 That sounds good, but it would add more moving parts. Might make sense i n some designs, but I'd prefer to avoid it. Maybe you're thinking of a moe bot with small eyeballs? If the eyeballs in a bigger one are quite huge and you can get the eyeball out first, then you would need to cut through some silicone, then getting the servos out should work as well. My point is the balls should be held only by the eyesocked (circle) as part of the skull, but if needed it could expand and form a bigger circle.
>>4335 After thinking about this a bit more, it came to mind that using a magnet connected to the servo and metal part to the eyeball, while the eyeball is also connected to a cable might not work. I'm concerned that the position of the eyeball couldn't be controlled very well. However, I tested it with a hollow flexible ball and a weak magnet (without servo). Now I can imagine it better and I'm more confident. However it might need a automatic re-adjustement mechanism. In case the servo looses connection to the eyeball it might move without the eyeball and then not know where it is.
>>4341 >That sounds good, but it would add more moving parts. This is a good example of the basic eye-control mechanism I have in mind: >>1657 (1st pic) The access mechanism would simply add a slide-out frame for the entire assembly to rest on. >then you would need to cut through some silicone, I would propose a design that had firm plastic 'frames' around the eye sockets, with a clearly-delineated seam that wouldn't require any special treatment to slide the eyes out for access.
Open file (432.91 KB 620x800 Nebula.png)
>>4351 Okay, that's different from mine, bc it will probably not be a seamless skin or plastic face on the outside. Which is okay, we need different approaches for every taste and use case. In such a case you might want to embrace the seams and color the face in different colors, bit like Marvels Nebula for example.
>>4355 Yeah, maybe a good idea. I'm not a art-designer per se, but I'll dabble around with different styles once my efforts have matured a bit.
Open file (129.78 KB 691x1037 IMG_20200702_020947.jpg)
Here's a lot of discussion towards moving eyes, following eyes and blinking eyes. https://dollforum.com/forum/viewtopic.php?f=6&t=103291 They're having a lot of trouble with space. Putting all kinds of stuff into a skull won't be easy. I thought about how to get the whole eye movement contraption smaller, thought the eyeballs would need to be bigger than human ones for anime eyes. One thing is, using servos as small as possible, not these bulky ones. If we can do it with normal dc motors then those would be even better. Of course, they are not so precise... I also thought I had a good idea, by putting one motor in the quite big eyeball. Problem is, I'd like to have the noise dampened. I also had the idea of using one motor for both eyes before, but human eyes can move independently. However, this does not apply for up/down movement, so we are down to three small motors then. The one in the middle would need to have an axis in both directions. Even better would be if the motors could be a bit deeper in the skull and also being used for something else (just a crude idea yet).
Open file (79.41 KB 1283x1064 IMG_20200701_210057.jpg)
I just listened to that interview here just by chance, while doing some chores: https://youtu.be/bnsgsPjILyQ - Very fascinating. How image recognition needs to work, so the system can think about it. It's one model that looks for different things in a image. It's inspired by neuroscience. Idea is that perception and cognition can't be disconnected from each other. Natural signals, segmentation and top down controllability are the keywords, the latter means for example when we're zooming into a picture in our minds.
>>4541 >Putting all kinds of stuff into a skull won't be easy. Yeah, I think that's mostly a misguided idea. Trying to bio-mimic absolutely everything that's been so elegantly designed into humans is, well, humanly impossible. :^) Better to keep most of the componentry safely protected inside the torso, etc, IMO. >>4777 >perception and cognition can't be disconnected from each other. Yeah, that's probably correct. Certainly it seems to ring true with some of the positions Carver Mead suggests for the field of Neuromorphic Computing he basically originated. Much of what we think of as 'cognition' is in fact neurological at a basic level instead of a higher level, and the perceptions are pushed as far out towards the extremity of sensory perception as possible in most the the higher life form's biological systems.
>>4786 IMHO at least one relevant computer should be in the head, to imitate humans. Also, stuff we have to put into the head isn't only that, but we'll need a lot of mechanisms in general there, so space matters. Think of facial expressions, microphones, speakers (mb in the throat), heating for the skin, tongue moving around while still leaving some space... Cleaning mechanisms... Okay, this is going OT towards >>9 (face/head general). Further discussion on what to put into the head maybe better there?
>>4792 Yup, all good points Anon.
Open file (43.59 KB 681x555 OAK-specs.jpeg)
Boards with cameras attached came up in the thread on SBCs, here: >>5705 OAK from OpenCV and a cam from Jevois where the computer is part of the camera. Fascinating, but might be a problem if one wants to put it in eyeballs and also make thouse water proof. OAK seems to be a bit big and the cams from Jevois have aircoolers... On the other hand for development my concerns might be irrelevant, since one can build something with them and replace them later with something smaller and cooler. The Jevois camera has shutter sensor with inertial measure unit and digital motion unit, gyroscope and all kind of sensors, wow: https://youtu.be/MFGpN_Vp7mg
Here's a video on eye movement. https://youtu.be/FaC2RXBss2c The human eye has six muscles, it can even roll sideways a bit. However, what always bothered me, is that so many fembot eyes can move independently up and down. I still think this isn't necessary. I'll look for a motor with two axes, for up and down movement.
>related xpost >>8659

Report/Delete/Moderation Forms

Captcha (required for reports)

no cookies?