Online path tracking with an integrated H∞ robust adaptive controller for a double-Ackermann steering robot for orchard waypoint navigation

Abstract

Navigation of agricultural mobile platforms in small-scale orchards poses challenges due to narrow row-end turning spaces and the need for precise path tracking in the presence of disturbances. The objective of this study is to improve path following and rapid turning maneuvers for a double-Ackermann steering robot by employing a simulation approach for PID-based waypoint following enhanced by learning-based H∞ robust adaptive control. With the zero-speed turning radius of the robot measured at 2.85 m, the primary question to address is determining the minimum achievable turning radius using the two controllers. For this purpose, a versatile framework for fine-tuning and analyzing of the controllers is presented in MATLAB Simulink blocks interfaced with the virtual replica of the robot in CoppeliaSim. A comparative study between the controllers is carried out involving three experiments: offline path following with a fixed number of predefined waypoints, online path following with continuously updated waypoints forming paths, and path tracking with disturbance rejection using the H∞ controller to reduce the radius of row-end turnings. Results indicate that while the PID controller achieves a minimum row-end turning radius of 3.0 m, the learning-based H∞ controller surpasses it with a minimum radius of 2.9 m. It is observed that a minimum of 4 waypoints is required for the PID controller to perform effective row-end turning in the offline experiment, with a higher number of waypoints enabling the robot to navigate through complex geometries and tight turns more effectively. Moreover, by incorporating an actor-critic structure, it has been demonstrated that the learning-based H∞ controller maintains stability even when facing wheel slippage disturbances, and outperforms the PID controller in online path tracking, particularly when maneuvering along a half-circle path. The framework proposed in this study contributes to improving autonomous navigation, particularly in determining the optimal number of waypoints and path configurations required for navigating agricultural robots with varying dimensions and steering mechanisms.