iRVision rotation

Hi all,

I made a simple place-part-in-box with 1mm clearance all around. BW CCD is fixed, box sits on a flat surface and Userframe is grid-calibrated. Rotation around the R-axis is clean, so TCP is fine. XY variations leave frame-offsets spot on, so Userframe is fine.

So how is it that offsets become up to 5mm off when box is rotated? I've tried turning orientation off and copying the reference every 30 degrees without changing anything else, no change in position: spot on. The problem begins when I turn orientation on, which seems to be the cause of my wrong-offset problem.

How should I go about fixing this?

Thanks in advance,

~ Tony

No, the position and rotation seem ok on-screen. The problem looks to be occurring during offset calculation.

Quote from Seanklein

How does you centre origin look in the GPM tool? Could be off centre which would explain an offset when rotation is enabled.

Could have been stray pixels unmasked which is throwing the centre off.

Could be, perhaps I should try a learning model to minimize this.

Quote from stare284

You may need to use the ADJ_OFS program.

I've heard of this method before but I'm not sure how to go about implementing it.

Quote from HawkME

I would recalibrate your camera. You you have positional error due to rotation that usually indicates the calibration was not done correctly.

Yes, I've tried several times and got consistent results. I've done similar work with a 7kg robot and haven't bumped into this before.

Quote from Fabian Munoz

ADJ_OFS does what any good common sense robot programmer does .

When a point is hard to teach "because it moves" , well, you teach the average (Known as "happy point" or "sweet spot") where the robot can reach it

What I do , I teach it normally and put the coordinates on an excel sheet, Then after few run you see the robot crashing and you teach again. After few teach, look at your excel sheet, find the average of the coord and punch them in. That's all you can do as a robot guy. The rest it's up to tool design, part moving, etc

I see. So basically you're taking the average number for each angle and finding the best fit curve. Cool!