Motion control of indirect-drive robots: Model based controller design and performance enhancement based on load-side sensors.

摘要

This dissertation is concerned with the motion control of indirect-drive robots. Indirect-drive robots use gears, such as harmonic drives and rotary vector (RV) reducers, for speed reduction and torque amplification. The gears introduce challenging control issues such as compliance, friction and transmission errors.;To account for the gear compliance, this dissertation presents the model-based design methods for the decentralized observer state feedback controllers for robots with joint flexibility. The design methods consider the MIMO nature of the robot dynamics, and the decentralized controllers are designed to ensure that the linearization of the nonlinear closed-loop system is stable. This guarantees the local stability of the nonlinear system. Furthermore, a closed-loop pole region is incorporated in the controller design to ensure certain local performance. The decentralized controller design problems can be cast as linear matrix inequality (LMI) problems. This dissertation further extends the controller design methods to the case in which the joint flexibility can be neglected in certain joints of the robot. In this case, the dimension of the controller design problem can be reduced.;To enhance the control performance, the benefits of load-side measurements in controller tuning based on experimental data are studied in the dissertation. In particular, the load-side acceleration measurement is utilized because of easy installation of MEMS accelerometers. The feedback and feedforward controllers are adjusted sequentially, and controller tuning is accomplished by the iterative feedback tuning method. The experimental procedures for the tuning of the feedback and feedforward controllers are provided in the dissertation, and the effectiveness of the load-side acceleration measurement in the controller tuning is demonstrated by experiments.;The third part of the dissertation is the investigation of the load-side measurements in the learning control of indirect-drive robots. It is demonstrated through analysis and experiments that the performance of the learning controller based on the load-side position measurement is superior to that of the learning controller based on the motor-side position measurement. Especially, the load-side learning controller is effective to compensate for the errors caused by the transmission error and friction of the gears. The dissertation further extends the study of the learning control based on the load-side position measurement to the case, in which the measurement is available only at sampling rates slower than the desired sampling rate of the learning controller. Two approaches are provided in the dissertation to deal with such learning control problems. The first approach uses a multirate kinematic Kalman smoother to obtain the load-side position estimate at the desired sampling rate by fusing the position measurement and additional load-side acceleration measurements at fast sampling rates. The second approach utilizes an interlacing technique to obtain the slow sampling rate position measurement, and updates the learning controller at the desired sampling rate using only the available slow sampling rate measurement. Finally, the effectiveness of the two approaches is demonstrated by experiments.