I feel like Fanuc should provide a good diagram or infographic on these frames.
Here is my understanding (which could be wrong, everyone who uses iRVision has figured out how to make sense of it in their own way):
Grid calibration frame = the user frame taught on the calibration grid. So this is the one you create when you place the grid calibration on a table, etc and touch off on the points of the grid to create a user frame. This allows the robot to know where the calib grid is when it does the calculations for calibration.
Offset frame = this is the frame in which the parts will be when you take the image. The offset values returned from iRVision are with respect to the origin of this frame.
From the iRVision training manual it explains the offset frame is the "frame in which the offset of the part is to take place".
Most of the time we put the grid calibration in the same place that we will be picking the parts. In which case, the grid and offset frame can be the same. However it doesn't have to be the same place. You can calibrate on one side of your workcell and then pick parts on the other side of the workcell. As long as you have user frames for each location and you set the appropriate value for Grid Calib Frame and Offset Frame.
Application frame = who knows? Even when I went to training, I felt like the teacher could not give a good explanation. The training manual only says that the Z of the camera should be normal to this plane. I usually use 0.