Neural Network Based Steering Controller for Vehicle Navigation on Sloping Land

Abstract This paper presents an autonomous guidance system of a wheeled tractor-like-robot on sloping terrain. A Neural Network (NN) vehicle model was developed for sloping land and trained using Back Propagation algorithm. Genetic algorithms were used to search the optimal steering values for different combinations of lateral and heading deviations. Using those values, a NN-based steering controller model was designed to generalize the optimal steering for different land-inclinations. Autonomous travel tests were conducted with a prototype test tractor along predetermined rectangular paths on sloping lands. It was found that the tractor could precisely follow the paths. The mean and standard deviation of the offsets along four linear directions of the rectangular path on 15° sloping land were 0.058 m and 0.063 m respectively, which are insignificant for tractor motion on agricultural farms.