Advice on sub-mm accuracy for robot arm
Hello,
I am implementing a full joint impedance controller on a factory robot arm (with feedforward coriolus, mass matrix, gravity comp, and friction comp). It works well, up to about 0.2 deg accuracy due to low gains (which I want) and static friction deadband. If I increase the coulomb friction compensation I introduce small high-frequency oscillations.
My ideas to solve this are either some secondary low amplitude damping term that allows me to increase friction comp and damp out its oscillations, or add some clipped/gated integral term. I was wondering if anyone has chased steady state error to ~0.01 deg and if either of my above ideas are standard in industry. Note I really just care about decreasing the error of my sensed versus realized joint angles, and I am not too concerned with hardware tolerances/physical accuracy. I more so just want my realized joint values to be within 0.01deg of my commanded, and I don’t want to increase my P gain. Any advice as to what’s happening under the hood on these sub-mm controllers would be greatly appreciated.