Optimal path generation for excavator with neural networks based soil models

Abstract

In order to automate the excavating process, the path of the excavator bucket tip should be optimally generated. The following four factors must be considered when the bucket path is determined: bucket volume (soil capacity in a bucket), reachability (backhoe structure limitation), time efficiency, and soil property. Among them, the soil property is hardly quantified due to the complexity of its mechanical behavior. This paper deals with a neural network model to identify the soil property. Human operator usually determines soil type by sensing its hardness given a specific path and then plans a safe and workable path. The neural network model proposed in this paper outputs the soil type with the force and trajectory inputs. The feasibility of the proposed system is proved through the experiments with a robot equipped with a force sensor.