Screw insertion using Robotiq Camera

Tyler_Berryman

Hi guys,

I was curious to see if I could use the vision system to find the through holes in a part. I tried teaching the M6 holes to the vision system, but I had to get very close to the part and this was limiting the field of view of the camera. So I decided to teach the entire part to the vision system, and I use an IF statement to teach several hole positions instead. I used a pallet function to pick up the 4 screws from the jig, and I used 4 IF statements to tell the robot where to insert the screws. In this demonstration, I am inserting the screws clockwise into the mechanical coupling. 

As you can see, the part can be moved around since the vision system is looking for the part after picking up each screw. 
*It would've been a good idea to use the aluminum fingertips without the silicone since the screws were sticking to the silicone!

Tyler_Berryman

Here is the program I used for the demonstration:
 Program
   BeforeStart
     parts≔0
   Robot Program
     MoveJ
       Waypoint_1
     Pallet
       Pattern: Line
         StartPos_1
         EndPos_1
       PalletSequence
         Approach_1
         Gripper Move 70%
         Waypoint_14
         Gripper Close
         Waypoint_15
     MoveJ
       Waypoint_1
     Camera Locate
       If parts≟0
         MoveL
           Waypoint_3
           Waypoint_2
           Gripper Open
           Waypoint_3
       If parts≟1
         MoveL
           Waypoint_5
           Waypoint_4
           Gripper Open
           Waypoint_5
       If parts≟2
         MoveL
           Waypoint_7
           Waypoint_6
           Gripper Open
           Waypoint_7
       If parts≟3
         MoveL
           Waypoint_9
           Waypoint_8
           Gripper Open
           Waypoint_9
           Halt
     Set parts=parts+1

Loïc

@Tyler_Berryman
That's an interesting application. In fact, I'm working on a similar application. My application need to be really precise and the bigger part will move between two insertions. I'm thinking about using first a camera to locate approximately where the hole is, and then insert the screw properly using a force sensor. Do you have an idea of the accuracy of a such system? Do you have any other idea for precise insertions? All ideas are welcomed.

Tyler_Berryman

@Loïc Would you be able to post a picture or a video of your current process? We do not have an exact specification on the precision of the vision system, but the gripping position identified using the vision system varies by about 3-4mm. In the video, I was placing an M6 screw into an M6 counterbored hole, so I did have a little bit of play. However, the use of the force sensor would've allowed me to guide the screw into the hole. 
Could you give me some information regarding the size of the part and the size of the holes? 

mathbelanger

WOW! Really impressive. Do you think that we could use this type of motion for precise dowle pins or with screws that need to enter into a very fit hole? I am looking to automate some assembly steps of our product. This would be a good way to do it. Let me know. 
Best ;-)

lakshmip001

I am using Robotic Camera to detect a circle of diameter in mm. Can we use Robotic Camera for detecting a hole of diameter 70mm? If not can I program it to detect 50mm?

Catherine_Bernier

@lakshmip001You can find all the details about the specifications of the camera like the smallest part that can be detected on our support website. 
Here is the page concerning the size of parts: http://support.robotiq.com/display/RVS/6.3+Vision+System+Specifications 
The minimum dimension detectable is 10% of the field of view, so 7.5 mm would be the smallest dimension the camera can detect.
This would work for the part dimension you have.

Sebastien

@Tyler_Berryman
Sweet application. One question: are the holes on the part through holes such that when you take a picture you have a good contrast in the hole (background compared to the parts since you see the background through the hole?)

Tyler_Berryman

@Sebastien Yes, the holes in the part are Counterbored holes. So when the camera takes a picture of the part, the background can be seen through the through holes in the part. The part used in the demo was the mechanical coupling that comes with the UR Kit for the 2F85 gripper. 
Here is a picture of the part from the Wrist Camera's point of view:

And here is a picture of the image template generated by the vision system :

As you can see, the through holes are included in the image template for the mechanical coupling. In my program, I was not actually locating the individual through holes. I was looking for the mechanical coupling itself, and I taught 4 different pick positions in the camera locate node. I figured that I would get a better detection score by looking for the part instead of an individual hole in the part.

lakshmip001

Catherine_Bernier Yeah I saw the specifications of camera. So I tried experimenting with the camera at different heights and with black circle on paper with various daimeter varying from 3cm, 2cm and 1cm. But it was able to detect only 3cm diameter circle. From the application given above I understood that there should be object which covers 10% FOV and then only it is able to detect 6mm circle. Is my conclusion correct. If not please advise me.

Thanks 

Catherine_Bernier

@lakshmip001 In order to see the smallest dimension you'll need to use the lowest position of the camera relative to your work plane, so about 7 cm for the UR5.

lakshmip001

@Catherine_Bernier i used the smallest dimension of 7cm from field of view. But still I am unable to detect a 2cm circle drawn on white paper witha marker. Please help me if I am anywhere wrong. And minimum Field of View is 10cmx7.5cm. Then Minimum object detected is 7.5cm^2. According to this when I calculated for the smallest circle. I got 3cm as the minimum radius. I tested experimentally to many times. But still couldnt get it. 

Warm Regards,
Lakshmi Prasanna.

Tyler_Berryman

@lakshmip001 There is not enough contrast between a black object and a white background. Here is a post that explains why : http://dof.robotiq.com/discussion/comment/1224. Essentially, white and black are different intensities of the same color. 

However, I did try teaching a black circle on a white piece of paper and it worked fine. Here is the setup that I used:

The camera is about 16cm above the object. Here is the detection score that I obtained with this snapshot position.

And here is the size of the black circle that I drew on the white piece of paper :

As you can see the black dot is about 7mm in diameter. 

The position according to the base is:
X = 347.04mm
Y = -309.92mm
Z = -4.70mm
Rx = 0.0242
Ry = -5.3594mm
Rz = 0.0173

Try this and let me know if it works!

lakshmip001

Thanks Tyler_Berryman  for the help. It is really useful. But I was unable to teach object using white paper with black circle. But I taught the the circle of 7mm using yellow background with blue circle. I could use the same to detect the b

lakshmip001

For my application I need a detection threshold of 40%. I am facing a vague object when the object is not present. Can I know how can I get rid of these vague object? .

Thanking you in advance.

David_Levasseur

lakshmip001, you might need to set your detection threshold a tad higher to prevent false detection. At 40%, the system can detect your object even if it not there. Look at the score you get when a false detection occurs during object testing and set your threshold level higher than this score.

lakshmip001

Hi David_Levasseur ,The object I need to detect is little reflective and thin sheet and I need to detect one object from four objects on the background. So for detection of one object from the four and then one from three, then one from two. It needs a threshold of 40%. And vague object detection is taking place at a detection threshold of 42%. I tried by increasing threshold. If I increase detection threshold, object is not detected when four are placed at a time.

Kindly please provide some suggestion.

Thanking you in advance.

Warm Regards,
Lakshmi Prasanna

David_Levasseur

Ultimately, if a part is too reflective to be taught, you can paint it or even try to teach a drawing of the part. You can also try this technique if you can't paint the part. You might also be interested by those 2 discussions about that matter:
http://dof.robotiq.com/discussion/441/teaching-a-shiny-part-to-the-vision-system
http://dof.robotiq.com/discussion/210/shiny-object-vision-challenge

Tyler_Berryman

@lakshmip001 Could you post a picture of your part? Or maybe a picture of the teach pendant when locating the part? This will help us get a better understanding of why your detection threshold is so weak for your parts. If you do not feel comfortable publishing pictures of your parts publicly, you can send to me in a private message!

Tyler_Berryman

@lakshmip001 As David mentioned, if you have a 2D CAD of the part, you can print it in a 1:1 scale an use it to teach the part to the vision system. Other than reflections created by the shiny part, this part should work fairly well with the Robotiq Wrist Camera. I would try using a 2D drawing of the part to teach the part to the vision system. Try printing the part on a colored piece of paper (Yellow would work well for this), this way you will have a high contrast between the black printed lines and the background. Once you have taken the 5 pictures, remove the piece of paper and put the real part in the camera's field of view for the scanning process.