*Please note, this is not a forum for talking about precision or repeatability. This is specifically for talking about accuracy.*
I'm about 3 years into an automation design job, and I'm now getting a feel for the accuracy limits of robots. Having come from a previous job doing more CNC-based designs, I have had to take a step back. For even a low-cost CNC, .003" accuracy is easily achievable. For a robot, .040" is common for us.
To get what we need with robots, we have to program in compensations found by trial and error. Again, I'm not talking about going back to the same position over and over... I'm talking accuracy, which means sending the robot to a place that it has never been before. This is necessary for picking scattered parts off a platform, using a camera to tell the position and orientation of the part. The parts are always in different locations, so the robot is constantly having to explore uncharted territory, so to speak. Yes, we have calibrated the camera. But it is becoming apparent that, even with a perfectly calibrated camera, robots lack accuracy. This means picking one dowel up .050" from the end, and picking the next one up .010" from the end.
All that is my motivation for posting this thread, but I don't want to get into specifics about my particular applications... I don't want to make it work... I don't want to fix it... I want to discuss the general accuracy limitations of robots. From what I can tell, most robot manufacturers do not state any accuracy rating for their robots, whatsoever. They all rate them on repeatability. I'd like to hear from the community, then, what is your experience with your robots?
Examples:
So, from my experience, we are seeing somewhere between .020" and .080" inaccuracy in robot picking applications where parts are scattered in random orientations. I wish I had known this going in; I may have done things differently; I may have planned on re-picking all parts; I may have never planned to rely on pickup accuracy in the first place; I might have used a gantry system with servos on linear rails and ball screws, with accuracy ratings, instead of using robots. Who knows!
Long story short, it would be nice to document some kind of understanding of what kind of accuracy can reasonably be expected of a robot. Maybe the industry will start actually measuring and stating this!
I'm about 3 years into an automation design job, and I'm now getting a feel for the accuracy limits of robots. Having come from a previous job doing more CNC-based designs, I have had to take a step back. For even a low-cost CNC, .003" accuracy is easily achievable. For a robot, .040" is common for us.
To get what we need with robots, we have to program in compensations found by trial and error. Again, I'm not talking about going back to the same position over and over... I'm talking accuracy, which means sending the robot to a place that it has never been before. This is necessary for picking scattered parts off a platform, using a camera to tell the position and orientation of the part. The parts are always in different locations, so the robot is constantly having to explore uncharted territory, so to speak. Yes, we have calibrated the camera. But it is becoming apparent that, even with a perfectly calibrated camera, robots lack accuracy. This means picking one dowel up .050" from the end, and picking the next one up .010" from the end.
All that is my motivation for posting this thread, but I don't want to get into specifics about my particular applications... I don't want to make it work... I don't want to fix it... I want to discuss the general accuracy limitations of robots. From what I can tell, most robot manufacturers do not state any accuracy rating for their robots, whatsoever. They all rate them on repeatability. I'd like to hear from the community, then, what is your experience with your robots?
Examples:
- We have a 6-axis Mitsubishi RV-2FR with a dual gripper. We have taught the tool center points (TCPs) correctly using best practices. We can jog the robot until a fine point indicator tip in the gripper at the TCP is exactly touching a fine point indicator tip mounted to a fixture on a platform. We can verify the tips are touching exactly using 2 microscopes. We can then rotate the robot around its TCP z axis by 180 degrees. If the robot were perfectly accurate, the 2 points would still be touching. But they are off by .020". If I then rotate around the x axis by 45 degrees, and try the experiment again, I might be off .040". This is all slow-jogging, and waiting for the servos to stabilize, so the effects of speed and acceleration should have no influence. The servos are going where they are told to go; I do not doubt the encoders. When we actually run it in a program (picking up parts), we can be off by more than .060".
- I have seen slightly better accuracy out of a UR5e 6-axis robot; however, this one is using a single gripper, so there are no compound angles. This may be helping us there. For that one I never see worse than roughly .020" error.
- On our Mitsubishi RH-3CH Scara, I do not see accuracy error at all; however, note that this robot never changes its orientation during pickup. All parts are picked at the same orientation (yaw angle about the z axis), so rotation errors are out of the question.
- We also have a Fanuc LR-Mate 200id/7L; I have not measured the accuracy, but I know that we have to compensate for inaccurate pickups by placing them in a fixture and re-picking them. This compensation can be up to .080".
So, from my experience, we are seeing somewhere between .020" and .080" inaccuracy in robot picking applications where parts are scattered in random orientations. I wish I had known this going in; I may have done things differently; I may have planned on re-picking all parts; I may have never planned to rely on pickup accuracy in the first place; I might have used a gantry system with servos on linear rails and ball screws, with accuracy ratings, instead of using robots. Who knows!
Long story short, it would be nice to document some kind of understanding of what kind of accuracy can reasonably be expected of a robot. Maybe the industry will start actually measuring and stating this!
Last edited: