# kuka positioning accuracy

• Hello
I need some clarification for robot positioning accuracy in general
Let's say this example
I jog the robot from P1 to P2, in World on X direction.
P1 x=200 y=0 z=0
P2 x=700 y=0 z=0
On actual position I read x=700mm but physicaly the TCP is located at 700.2 mm (I used a precise caliper rule ). That 0.2 mm is error for position accuracy?
Thank you!

• Articulated arms are generally not rated in accuracy, but in repeatability. It will go to the same spot every time with great repeatability, but internal measurements may be off. I think KUKA offers their robots in an HA (high accuracy) variant. All this means is that they measure the arm very carefully so that its kinematic model is accurate to that specific arm.

• Articulated robot arms have extremely complex (to the point of being non-modelable) issues with axis backlash and rigidity. By comparison, CNC machines are dead simple.

There are, however, "best practices" that can minimize these issues over small areas and short distances. For example, your 500mm move can probably be made much more accurate by incorporating anti-backlash motions at each end. This does not make the robot more accurate, but by pre-biasing all the axes' backlash (or, more formally correct, "lost motion") in the same directions, the effects can be minimized.

Something basically like this:

Code
``````PTP P1
PTP_REL {A1 0.1, A2 0.1, A3 0.1, A4 0.1, A5 0.1, A6 0.1}
PTP P1 ; "pre-biased" start position
PTP P2
PTP_REL {A1 0.1, A2 0.1, A3 0.1, A4 0.1, A5 0.1, A6 0.1}
PTP P2 ; "pre-biased" end position``````

(this is a crude example. You can also do this in Cartesian space, or perform your anti-backlash motion as part of a pre-calculated entry/exit path at each point)

If you measure the robot's motion distance between the pre-biased points, you should find that the distance is much closer to the nominal. Using this sort of method, I've achieved measurable motion accuracies down in the 10-micron range, over short distances.

There are complicating factors, however. If, for example, one of your axes passes through a critical point between P1 and P2, that will cause unpredictable accuracy errors. For example, if A2 is -100 at P1, and -80 at P2, that means that the force of gravity (and the counterbalance actuator) will be pulling A2 in opposite directions at each of these points, which will fight against your anti-backlash techniques. Orientation changes to the tool will also act against your anti-backlash techniques.

• Hello

SkyeFire

I never heard about it, I can explain precisely what it is anti-backlash?

Quote

incorporating anti-backlash motions

You can write something about a series of robots HA (HIGH ACCURACY)?They are suitable for precise milling models?Are to eliminate the backlash and vibration in the robot?whether these robots are comparable with CNC milling machine?

I have application to pick parts from the table with many symmetric position . I used first picking position and I created a matrix using While instruction for rows and columns.
I added +50 mm for every new position. But at the end where the position is 500/500 I have a deviation around 0.3mm on Y. I know that this error came also from my base but on X deviation 0.1 mm. And Z is also not correct when the robot is far from robot root (from 0 to 1,5 mm when Y=500 and X=0 to 500). Is an Agilus robot

I have application to pick parts from the table with many symmetric position . I used first picking position and I created a matrix using While instruction for rows and columns.
I added +50 mm for every new position. But at the end where the position is 500/500 I have a deviation around 0.3mm on Y. I know that this error came also from my base but on X deviation 0.1 mm. And Z is also not correct when the robot is far from robot root (from 0 to 1,5 mm when Y=500 and X=0 to 500). Is an Agilus robot

Yeah, that's not unusual. If you're not using anti-backlash procedures, these numbers aren't too bad. Probably a portion of that error is caused by misalignment of the Base frame -- if you take enough samples across the robot's working volume, you can average out that error by adjusting your Base ABC values.

• No articulated robot will ever be comparable with a CNC machine within the current state of the art. Period.
The HA-series robots reduce backlash by using tighter build tolerances, but the kinematic style of any articulated robot arm simply makes matching a CNC machine impossible. Robots don't simply have much more complex kinematics, they're also far less rigid -- CNC machines use massive amounts of framing and "enclosed" kinematics to make themselves effectively immune to gravity effects and milling forces. Robots are much lighter and have "external" kinematics.
("internal" kinematics in this instance refers to how any CNC only works "inside" itself -- the tool never moves outside the framework/bracing of a typical CNC machine, even a Knee-mill type. Robots, OTOH, are always working 'externally' to their mounting base)

For discrete-point operations, anti-backlash procedures can help, over short distances. But for milling-style operations, there is less they can do. Robots are inherently less accurate along a path than at motion endpoints. Still, for milling operations, things like avoiding joint reversals, running "layered" cuts in the same direction rather than alternating, etc, can help reduce (but never eliminate) some of the the issues. If you have the time, it's possible to manually "tune" a milling path over multiple iterations until it's nearly perfect -- that's effectively turning the problem from one of accuracy to one of repeatability, which robots are much better at (although still not as good as a CNC machine).