Is it possible to offset snapshot positions after calibration?

Grady_Turner

I have seen a need for this several times.  An example would be a pallet where the nests do not perfectly locate the parts.  I have used the robot to iterate through the pallet and take a picture at each pallet position to find the part in the nest.

Is this possible (or could it be possible) to do with the RQ camera kit?

Alexandre_Pare

@Grady_Turner
This is a good question and I got a similar application where we would need to empty various trays that are stacked one of top of the other. I will have to run some tests on this. This will basically depend on how the math is handled in the camera software. The technique that I am going to try Monday is to offset the snapshot position by a fixed amount every time the code runs through this node, let's say 20mm in a specific direction. And I will share results here. The snapshot position would be the waypoint that is just above the camera locate node.

PierreOlivier_Proulx

@Grady_Turner As it is now, the vision server supports only slight translations on the snapshot position. Let me know what are the results of your tests.

However, I added something in the Camera Locate that makes it possible to offset not only the snapshot position but also the calibration plane. With that modification I was able to unload a stack of three trays. Basically, the plane on which the part detection was made was indexed with the stack height.

I'm short on time but I plan on sending a video of this.

PierreOlivier_Proulx

It is something you will achieve through scripting. In short, hidden in the Camera Locate node, there is variable named snapshot_position_offset that is initialized to p[0,0,0,0,0,0]. If you set this variable, it will offset the calibration plane. You have then to manually offset the snapshot position (tray_var in that case) and move to the new snapshot position. With this example, the teaching was made at the height of the first tray.

Grady_Turner

@Alexandre_Pare that was my idea, the issue I see is the possibility of "matching" the model with some random objects in the background (like the trays).  But I will do some testing as well; let me know what you find!

Grady_Turner

@PierreOlivier_Proulx sounds great!  I'd be very interested in having the functionality.  Is that something you added to the URCap, or something I can add in Polyscope?  (can I simply offset my snapshot pose variable?)

Grady_Turner

@PierreOlivier_Proulx  Thanks!  I will do some testing with this.  This really adds a lot of flexibility!

PierreOlivier_Proulx

@Grady_Turner That feature is not available on your version. I will send you an update shortly.