Adaptive neural network-based path tracking control for autonomous combine harvester with input saturation

Abstract

Purpose

To reduce the effect of parameter uncertainties and input saturation on path tracking control for autonomous combine harvester, a path tracking controller is proposed, which integrates an adaptive neural network estimator and a saturation-aided system.

Design/methodology/approach

First, to analyze and compensate the influence of external factors, the vehicle model is established combining a dynamic model and a kinematic model. Meanwhile, to make the model simple, a comprehensive error is used, weighting heading error and position error simultaneously. Second, an adaptive neural network estimator is presented to calculate uncertain parameters which eventually improve the dynamic model. Then, the path tracking controller based on the improved dynamic model is designed by using the backstepping method, and its stability is proved by the Lyapunov theorem. Third, to mitigate round-trip operation of the actuator due to input saturation, a saturation-aided variable is presented during the control design process.

Findings

To verify the tracking accuracy and environmental adaptability of the proposed controller, numerical simulations are carried out under three different cases, and field experiments are performed in harvesting wheat and paddy. The experimental results demonstrate the tracking errors of the proposed controller that are reduced by more than 28% with contrast to the conventional controllers.

Originality/value

An adaptive neural network-based path tracking control is proposed, which considers both parameter uncertainties and input saturation. As far as we know, this is the first time a path tracking controller is specifically designed for the combine harvester with full consideration of working characteristics.

Citation

Zhang, Y., Wang, L. and Liu, Y. (2021), "Adaptive neural network-based path tracking control for autonomous combine harvester with input saturation", Industrial Robot, Vol. 48 No. 4, pp. 510-522. https://doi.org/10.1108/IR-10-2020-0231