IJCAS

This paper introduces a novel control approach for automated excavators combining model-based and learning-based techniques to enhance control accuracy. The feedback linearization technique is employed based on error dynamics in designing boom and bucket velocity controllers incorporating the driver’s manual arm control. Additionally, supervised learning is used to approximate inverse hydraulic actuation system and to compute joy stick control inputs corresponding to the desired control velocity. To further refine control precision reinforcement learning is used to optimize the driver’s manual arm manipulation within a given cycle time. The performance of the proposed methodology is demonstrated through simulations on a 30-ton excavator and compared with results based on model-based techniques.

Keywords: Excavator, feedback linearization, intervention, inverse hydraulic actuation, reinforcement learning, semi-automatic grading, supervised learning.