Control design for robust performance of a direct-drive robot

Abstract

A b s t r a c C A n experimental approach to achieve robust performance of direct-drive robot motion control i s presented in this paper. I t consists of: (i) decoupling the robot dynamics via feedback linearisation; (ii) frequency domain identification o f the decoupled dynamics; (iir) compensation of these decoupled dynamics using feedback controllers designed via p-synthesis. The designed controllers ensure robust performance, i.e., guaranteed accuracy o f robot motions despite uncertainty in its dynamics and disturbances affecting the robot operation. Theoretical aspects of the control design are formulated. I t s practical implementation on a direct-drive robotic arm is demonstrated in detail. Experimental investigation confirms the quality o f the design: specifications on performance and robustness are practically realized.

Discrete-time sliding mode control of a direct-drive robot manipulator

“…It appears that the DSMC of the RRR robot does not give the performance achievable with H ∞ feedback control [15,16]. This could be expected, since H ∞ controllers are based on models with flexibilities included, while the DSMCs consider only the rigid-body dynamics that remains after feedback linearization.…”

“…Practical procedures to identify an FRF as shown in Fig. 3 are explained in [15,16]. Note that the given FRF represents the dynamics of joint 1, as it was obtained by averaging the dynamics identified for different robot postures [16].…”

“…2, is an experimental facility for research in motion control [13][14][15][16]. Closed-form models of the robot kinematics and rigidbody dynamics are available in [14].…”

“…These jumps are due to oscillations observed in the error signals. The oscillations correspond to severe peaking around 20 Hz in the sensitivity function [15][16][17], that can be computed using the identified FRF shown in Fig. 3 and the equivalent PD controller (21) with λ p = 5.…”

“…For β 1 ≤12, the relative damping is unreasonably high, and may excite flexible dynamics and amplify quantization noise. An appropriate set of PD gains, away from the "stability" boundary β 1 =12, can be found using loop-shaping, as explained in [15][16][17] ). Experiments showed no need for higher ρ i .…”