Eye Mechanism Assembly

This new mecanism is inspired from Mats Onnerby and Bob Houston. It replaces my previous eye mecanism which you still can find after the below tutorial.

Download STL from the Gallery

You will need to print all these parts at a good resolution:

  • 1x2xEyeBallFullV2
  • 1xEyeHingeCurve
  • 2xEyeHinge
  • 1xEyeHolder
  • 1xEyePlateLeft
  • 1xEyePlateRight
  • 1x EyeSupport (print with support)
  • 1xEyeTo Nose

Depending on your camera choice, print:

  • EyeBallSupportHerculeLeft
  • EyeBallSupportHerculeRight
  • EyeBallSupportLifeCamHDLeft
  • EyeBallSupportLifeCamHDRight

I used 2 DS929HV servos from Hobbyking because they are stronger, but you also can use some DS928HV or even some cheap SG92R or HXT900.

You can use two sorts of webcam:

  • Hercules Twist HD webcam
  • LifeCam 3000 HD Microsoft (possibly 5000 as well if the PCB is the same)

Dismantling the LifeCam 3000 HD: https://astrophotovideo.wordpress.com/adapting-a-webcam-to-a-telescope/

image(7)

A video tutorial in French is available, you can have subtitles in English through the settings of YouTube:

 

Step1:

DSC07217

Redrill all the holes on EyeToNose with a 2,5mm drill.

DSC07218DSC07221

Tab all holes with a tab of 3mm diameter.

DSC07219 DSC07220

Glue, with acetone if you have ABS prints or Epoxy for PLA, EyeToNose to EyeGlass.

DSC07222

Tab after redrilling the 4 big holes of EyeSupport with a 3mm tab.

DSC07223

Redrill with a 3mm drill the two holes of EyeHolder.

DSC07224

Insert and mount four screws of 3mm. You can add bolts as shown to secure them.

DSC07225

After redrilling the holes of both EyeBallSupport with a 3mm drill, mount them as shown. At this stage you can also add the the Hercules camera on the EyeBallSupport. My picture does not show the mounting of the cameras HerculesHD because I was trying a different sort of camera at the time.

IMG_2688Here, I am using a LifeCamHD 3000 from Microsoft. The image quality seems less good than the Hercules Twist HD. Notice that the EyeBallSupport is different because the PCB of the camera is set horizontably.

Step2:

DSC07227

Mount the one DS929HV servos from Hobbyking with its screws.

DSC07228

DSC07228

Set the servo at 0 degree with the horn set as shown.

DSC07229

Mount the second servo DS929HV.

DSC07230

Set the servo at 0 degree and mount the horn in a position similar to a 45degree  angle.

DSC07231

Add EyeHolder with two screws of 3mm diameter.

DSC07233

And fix the mecanism to EyetoNose. Once mounted it should be able to rotate freely on its shaft.

DSC07234 DSC07235

Mount and screw the EyeHingeCurve between the servo horn and EyeToNose. Again the mecanism should be able to rotate freely on it’s shaft.

DSC07226

DSC07237

Add on the back of EyeBallSupport the EyePlates. You can either glue them or screw them.

DSC07238

DSC07239

Mount and screw the two EyeHinge between the top servo and the Eyeplates.

Step3:

Now we are going to create some realistic eyes with a few simple steps.

DSC07233

Unmount the mecanism by unscrewing the two screws of EyeToNose to have access to the front of the eyes.

IMG_2689This is a full EyeBall set.

Bellow is the ring order placement:

IMG_2690IMG_2691IMG_2692IMG_2693IMG_2694IMG_2695

Now let your creativity get wild!!

IMG_2697Spray paint or use a colored marker of your choice to create a base on the pupile ring. In my case I used mat blue spray.

IMG_2709Make some little lines with another thin black marker, from center to the outside. Perfection is not required as you can see on my picture. But if you are good and creative, you can really obtain something very realistic.

IMG_2698Add some little lines with your thin marker on this outer ring as well.

IMG_2699Spray paint in black mat finish the iris ring.

IMG_2700Now assemble all the rings together.

photo(10)You can also replace the outer ring, pupile, and iris with a Fish Eye Lens for Smartphone. But it looks less realistic.

IMG_2703To make the eyes even more realistic lets be creative!! Now we are going to create the eye transparent cover.

IMG_2705You will need a heat gun, a glass ball of approximately 3 cm diameter (I used the perfume cap “Dior, J’adore” of my wife, fancy eyes!!  You also need a piece of thermoformable plastic (cristal clear is better, other wise it will alter the camera vision) I personnaly used a piece of a light bulb blister. But any cristal blister will do the job.

IMG_2706Clean up all the surfaces to make sure you won’t alter the transparent plastic with dust or scratchs during thermoforming.

IMG_2707Apply hot air on the plastic and when it starts to become soft, apply and stretch it over the glass ball. Wait for cooling.

IMG_2708Mark the circle using the pupile ring, and cut out with scissors the lentille.

IMG_2701IMG_2702Add a bit of glue on the perimeter, carefull to be clean and avoid finger traces on the inside surface.

IMG_2704Glue the lentille on the pupile ring. Let it dry.

Do the same with the second eye.

image(7)

Mount back the mecanism on the EyeGlass. There you go, you have an impressive looking InMoov!!!

DSC07240

You are now ready to install the whole face and eye mecanism into the head. See the Hardware map for to check your default servo positions. Make sure no cables are blocking the eye mecanism to rotate in X and Y directions.

 

 

 

Tutorial for the old eye mecanism:

My original STL files are still available here.

I used 3 DS929HV servos from Hobbyking. The two servos mounted for the right/left movement are connected with Y connector, so they receive the same data and act simulteanously.
Some of the pictures in this tuto are showing parts that might be a bit different than the one you actually printed. This is due to different iterations.

Start by screwing the two servos to EyeSupport. In the tuto I have attached only one, but it is best to have two.
Set your servos at 90 degrees using your Arduino.
Attache the actuators of your servos in this position.
Mount EyeCamera part and make sure it can rotate freely on the screw.
Mount the second EyeCamera part
Now add and screw the third servo to EyeToNose
Your part will look a bit different in length.
Be sure to have your servo set at 90 degrees and mount the actuator this way.
Attach EyeToNose to the EyeGlass part
This is a bit difficult because the access with the screw driver isn’t straight.

Now fix EyeMoverSide through EyeCamera to the actuator of the servo. You can also mount a second servo instead of using EyeMoverSide.

two servos makes the X movements more sturdy.

Repeat this on bothe eyes.
Mount EyeMoverUp, it should NOT be tight screwed
Fix this assembly to EyeToNose, going through the eye space is an option to make it easier.
Fix the EyeMoverUp to the bottom servo actuator.
Normaly your mechanism should be ready for movements.
Making test with your Arduino is a good thing at this point. Be sure to start with small range degree movements, especially for the up and down movement. You can start with 30 to 120, it should be fine.
What follows is what I did to make the cameras look more like eyes.
I used a ping pong ball cutted with a knife and small scissors.
The hole where you insert the lens can be rounded up with sand paper, because cutting a clean hole with a knife isn’t easy.
Adjust the hole precisely to your lens and PCB board
I used a piece of heavy tape to cover the back of the PCB, for two reasons:
First to avoid using hot glue directly on the components and also to shutter the back of the lens to keep it dark (Oddly light coming from the back of the lens interfer with the image received by the camera.)
Now you can mount it to the head. You can see here my two different cameras.
On the left the Megapixel and on the right the Hercule twist.
The connections of these three servos will be added on the Arduino board, once the InMoov service will be implemented for that.

Comments 24

  1. Profile photo of juerg

    Hi Gaël
    Tried to advance to the new eye mechanism (2015). Small problem with the eye hinges for the X-movement. I measure the distance between the cam holders turning axis as 54 mm. The eye hinges you say to use twice only add up to 2*26=52 mm and are not taking the angle to the servo arm into account. This gives me a cross-eyed result. Would it not be better to connect the 2 cam holders over a long straight connector to keep them in parallel and connect only 1 side to the servo?
    Or maybe I wasn’t able to follow your detailed instructions correctly :-(
    Regards
    Juerg

    • Profile photo of Gael Langevin

      Hello Juerg,
      Something must not be mounted correctly, I suspect.
      I designed to specialy avoid cross eyed effect, in fact the eyes should be having the opposite effect for the purpose.
      The two cameras should not be not aligned.
      My previous eye design had a unique connector for both eyes, and most people would align the cameras to have a parallel view which gave a crossed eye gaze.
      Could you make a thread on the InMoov forum and post pictures? It will be easier to understand what went wrong, and if I need to review something.

  2. Profile photo of Scott Hodges

    Hi Gael,

    I have 3 questions about the 2.0 version of your eye mechanism. First, I noticed that the X axis pivot points are slightly off center. So the left lens/eye is a little closer to the left eye hole when looking right and the right eye is closer to the right eye hole when looking left. Is there a mechanical reason for the offset? Clearance perhaps? Is it noticeable?
    The second question is a little harder. Human eyes have over 135 degrees of vision but can only focus within about 6 degrees. So they move around a LOT in order to focus on anything within that 135 degree plus arc. (They are called Saccades, I think.)
    Here is the question. Do the inMoov eyes move strictly for the aesthetics of eye movement, or is there some true object-tracking/face-tracking computational easier-than-moving-the-whole-head logic in MRL someplace? I ask because I have tried face-tracking with a 3-axis head and 2-axis eyes on another platform and it was extremely difficult to make work using both head and eye movement. Can MRL do that or does it effectively turn off eye movement while tracking?
    Last question. Given that you are using an HD Twist in your eye, what do you think the degrees of vision is in your bot? How much does eye movement improve that range of vision?
    Sorry for the long comments/questions. Thanks in advance for your response.

    • Profile photo of Gael Langevin

      Hi Scott,
      The off center is on purpose, it is to avoid crossed eye effect on the robot. Too many people using my previous eye version, didn’t care much of that effect and they wouldn’t adjust it correctly, but it bothers me very much to see pictures of InMoov with crossed eye effect.
      Unfortunately Alan Timm has modified that again and the eyes look again crossed eye.
      Cameras don’t care if the vision isn’t aligned because we can set within MyRobotLab the middle point where ever we want.
      MyRobotLab has a 4 PID tracking system which is specially implemented for InMoov’s vision. The eyes move faster than the head when searching for a face or an object. Yoy can adjust the speeds of the PID in the python script. The 4 PID has been developped by Alessandro Didonna, who’s often on MRL.
      The HD twist doesn’t have a very wide range of vision, therefore, it is possible to have one eye with a HD twst standard, and on the second eye to add an 160 degree fish eye lens for iphone. Kevin watters has done it and it seems to work pretty well.
      Remember that Windows cannot (as far as I know) mount two cameras at the same time, a pitty for InMoov.

      • Profile photo of Scott Hodges

        Thanks for the quick response. I suspected maybe the offset pivots had something to do with solving the cross-eyed problem. I have 3 interesting eye versions now and did notice that Alan’s camera holders are not set up for the twist, so it looks like I’m either going to use your design or maybe do a little remixing.
        ….Windows can’t handle two cameras…..OK, that’s interesting. I’m guessing you mean that there is only one video channel via USB. I didn’t even stop to think that even the Oculus Rift only has one camera and that is for positioning, not real-time stereo viewing. I know you and the MRL folks are way ahead of me, but I’ll talk to some colleagues to see if they have any ideas.
        So far, although I’ve been printing for weeks, I do not have any complete subsystems ready to go yet. I’ve been testing servos and hands with other control tools I have, so I haven’t really jumped in to see MRL’s capabilities. Glad to hear it uses all servos for tracking. Really looking forward to making mine do that.
        Thanks again for the answers.

        Scott

  3. Profile photo of Scott Hodges

    Hi Gael,

    I LOVE the new clear eye covers! Your technique for creating the plastic cover is awesome and will probably work well regardless of which of the dozen or so eyeballs is used. Fortunately (great minds…wives of great minds….) my wife also wears J ‘Adore so I will try this as soon as I get my lab up and running again. (Been sidetracked with a couple of other projects…)

    Thanks!

    Scott

    • Profile photo of Gael Langevin

      Lucky you are to have a wife with good taste!
      Yes I think the plastic cover technique can work on most eyeballs. What is important is to make the diameter of the cover the same size of the Iris, this avoids to show the overlap of the cover on the white of the eyeball.

  4. Profile photo of Pascal

    I builds inMoov head for my two younger childs as a holiday activity since ten days. I made eyes mechanism and eye balls with eye lens effect with a transparent sheet curved properly. This is my technic without glass sphere.

    I used a 4cm diameter ping-pong ball as a support instead of 3cm diameter glass sphere. I was afraid that the heat could melt the ping-pong ball; but it was not the case.

    I used for the clear transparent surface to be thermoformed a transparent sheet to protect and cover bookbinding documents, the one you can buy in a stationery.

    I used the hair dryer on position max heating and put it nearby the surface of the transparent sheet (3cm approximately), the sheet being laid on the ping-pong ball put on the table. I used weights on the edges of the transparent sheet to maintain it outstretched on the ping-pong ball put on the table.

    When heating enough the transparent sheet it begins to go down to the ping-pong ball and become a little bit adhesive to it, but not on all the top surface needed for the eye. Then, I suddenly catch the edges transparent sheet with hands (somebody else directing the hair dryer in another direction so that I don’t burn my hands), and I stretch out the transparent sheet along the round ball surface and keep it stretched the time it cools down and become solid again.

    Then I have nearly the third surface of ping-pong ball covered with the transparent sheet following exactlyt the round surface of the ping-pong ball. I can then put the iris ring on it, draw the circle of the correct size and cut. I have very nice eye lens with this technic and the ping-pong ball never melt nor deformed with the heat; it is made of a material very heat resistant.

    So you have a technic with cheap ping-pong ball. What is more, the 4cm diamater of the ping-pong ball is more suitable for the eye lens curvature to be in harmony with the printed eye ball of inMoov.

    The final look of eyeball is nice (and it works weel with the webcam to see clear pictures, I mounted Hercules Twisted HD cam bought new for 8€ = $10 on the net with cdiscount and tested it on the eyeball with eye lens) . I just am not great in drawing and cutting a perfect circle on the curved transparent sheet after thermoforming so I don’t have perfect match of the edges of the eye lens with the edges of the iris on my eyeball when I glued it, I can improve!

Leave a Reply