Simulation environment development and control for path tracking in greenhouse harvesting and crop transportation with mobile robots

Abstract

This research presents a trajectory tracking methodology designed for differential drive mobile robots (DDMR), specifically focused on agricultural robotics applica- tions. The study encompasses a detailed exploration of the kinematic model of these robots, both for unicycle models and robots with Mecanum wheels, as well as a dy- namic analysis of them. Additionally, the design and validation of control techniques such as Proportional-Integral-Derivative (PID) control for Direct Current (DC) mo- tors, Model Predictive Control (MPC) with nonlinear feedback based on Lyapunov, and linear feedback control are carried out. The validation and integration between the designed control systems and the kinematic models are performed in two de- signed simulated environments, using the Robot Operating System (ROS) and a Python-based environment. This approach allows for perfect coordination of the system, incorporating trajectory generation using ArUco markers along with vari- ous trajectory planning strategies such as A* (A Star), Rapidly-exploring Random Trees (RRT), and Dynamic Window Approach (DWA), in order to generate work routes within the agricultural work environment, in our case study, the methodology is applied within a greenhouse setting. This strategy considers various operational phases, including initiation, harvesting, and the transportation of produce to the designated storage area. Python environment works as a preliminary interface to validate and adjust the parameters of the controllers and algorithms for optimization and trajectory gen- eration, while Gazebo and ROS integrate seamlessly to create a cohesive system architecture that allows validating the behavior of the designed model under real conditions. The results show the practical feasibility of the proposed methodology, demonstrating good accuracy in trajectory tracking both in the simulations within the greenhouse environment and in the validation in the Gazebo environment. The maximum average Mean Squared Error (MSE) was 1.68 × 10−4, and the average Integral of Absolute Error (IAE) was 4.25×10−3 for the trajectories within the sim- ulated environment in Python for the greenhouse. Meanwhile, for the environment in Gazebo, the maximum average Mean Squared Error (MSE) was 1.68 × 10−4, and the average Integral of Absolute Error (IAE) was 4.25 × 10−3. A notable aspect of this approach is its implementation in a specially developed virtual environment. The simulation in a controlled environment not only offers a platform for testing and refining algorithms in a safe context but also provides the opportunity to anticipate and resolve potential challenges before their application in real-world settings.