Online learning control of hydraulic excavators based on echo state networks

Abstract

In part of recent advances in automation of construction equipment, much research has been conducted on the control of hydraulic excavators in both industry and academia for the benefit of safety and efficiency. However, most relevant work have employed model-based control approaches that require a mathematical representation of the target plant. For hydraulic excavators, obtaining a useful dynamic model for control can be challenging due to the nonlinear nature of fluid dynamics. Furthermore, the underlying hydraulic circuit of an excavator may vary depending on the specific type and manufacturer of the excavator. With this in mind, this thesis investigates the feasibility of an online learning control framework to the position control of hydraulic excavators. The online learning control framework that is utilized in this work is based on echo state networks (ESNs). Two ESNs are deployed within the control framework: one is used to learn the inverse dynamics of the plant, and the other to generate the control input pattern in order to achieve the desired trajectory. Without prior knowledge of the plant, the control framework learns the input patterns, based solely on the input and output sequence of the plant, which drive the output of the plant to the desired trajectory. While such framework has been studied before, here, modifications are proposed to the previous control framework such as employing the receding horizon principle and using derivative information of the plant. It is shown that the modifications enhance the overall performance of the controller while maintaining the computational complexity at the same order of magnitude. Two strategies are investigated for the position control of hydraulic excavators. Within the first strategy, the online learning control framework serves as an inner-loop control of the hydraulic servo system. An outer-loop control based on impedance and sliding mode control is used to generate desired cylinder forces. The integrated inner-loop and outer-loop controller has been simulated on a hydraulic excavator simulation environment. In the second strategy, the online learning control framework is directly applied to the position control of the joint angles. Only joint angles are measured for feedback and remote control valve (RCV) pressures that correspond to the plant input is generated directly. This strategy has been experimented on a 21-ton class industrial excavator.