Automated loading of a hydraulic excavator using nonlinear model predictive control with preference-based calibration

Abstract

This study deals with the control methods for automating the operation of a hydraulic excavator loading soil onto the bed of a dump truck. The loading operation requires moving the bucket to a target position without contacting the dump truck. In addition, there is a demand to minimize soil spillage during the loading process. We applied Model Predictive Control (MPC) using constraints of the hydraulic excavator and objective functions corresponding to the requirements of the loading problem. To achieve the desired behaviour using MPC, the weights of the objective function need to be tuned appropriately. Tuning these weights is not an easy task, even for control engineers. To solve this problem, we utilize a weight tuning method based on preference learning. Using this tuning method, the weights can be tuned by repeating the process of selecting the result that the user finds preferable from pairwise tuning results. To confirm the effectiveness of the proposed method, weight tuning experiments were conducted in a simulation environment with multiple users with different levels of skill and knowledge. Through these experiments, we confirmed that each user was able to tune the weights to achieve a desired loading behaviour that appropriately reflected their preferences, independent of their levels of skill and knowledge.

Keywords

• model predictive control
• automated excavator
• preference-based optimization