Vision-Based Nonlinear Control for Furrow Following Using Convolutional Neural Networks

An omnidirectional mobile robot equipped with an ASUS Xtion sensor for image acquisition and an Orange Pi 5 as the central computing unit running Ubuntu 22.04 and ROS 2 was used as the experimental platform. The purpose of the robot is to collect RGB image data from an agricultural furrow while driving, as well as the corresponding control commands, including velocity given as a PWM (pulse-width modulation) signal and steering commands expressed in degrees. The main objective is to use this dataset to train a CNN–FC (Convolutional Neural Network–Fully Connected) model to achieve vision-based nonlinear control for furrow following in agricultural automation.