Autonomous Navigation in Orchard Rows: A Vision-Based Local Motion Planner Exploiting Smooth Transition Functions

Autonomous navigation in agricultural applications presents several challenges in perception and control due to the environment’s variability and irregularity. This paper introduces a novel approach to address this task, relying only on local data and without pre-defined paths. The proposed system enables the robot to execute two key manoeuvres: maintaining straight-line navigation within rows and performing precise turns at row ends. A unified control law, driven by a computer vision system, dynamically switches between these behaviours, exploiting a smooth transition function inspired by bumpless transfer control techniques. This method ensures a continuous control signal, achieving robust navigation in variable and dynamic agricultural setting. Simulation results and real-world validations are provided to demonstrate the effectiveness and performance of the proposed control strategy.