Investigation of robust path tracking, collision avoidance, and teleoperation for assisted navigation of an agricultural mobile robot in unstructured environments

The integration of autonomous navigation, collision avoidance, and wireless communication for teleoperation presents an opportunity to improve the reliability, safety, and adaptability of agricultural mobile robots operating in unstructured field environments. During actual navigation in dynamic and unstructured orchards with variations in terrain, mobile platforms often exhibit nonlinear dynamics, making the design of an effective navigation controller challenging, particularly when traveling on uneven terrain, slopes, bumps, and in the presence of nonholonomic constraints. Effective navigation in small-scale orchards with narrow row spacing necessitates precise path tracking and row-end turning maneuvers. The first part of this study was motivated by the need for a simulation framework to design and experiment with different path-following control strategies to be implemented on agricultural mobile platforms of varying sizes and steering mechanisms. The aim was to evaluate their performance in accurate path following and determine the minimum space required for row-end turnings. Therefore, the main objective of the first study was to design and compare two controllers for guiding a simulated double-Ackermann steering mobile platform along various waypoints and reference paths within a simulated orchard environment with narrow row-ends. Simulation experiments demonstrated that the learning-based H-infinity controller significantly outperformed the PID controller, achieving a minimum row-end turning radius of 2.9 m, compared to 3.0 m for the PID controller. The H-infinity controller exhibited superior stability under wheel slippage disturbances and was particularly effective in online path tracking scenarios, such as navigating tight half-circle paths. The proposed framework, implemented in MATLAB Simulink and CoppeliaSim, provided a versatile platform for optimizing control strategies and configurations, enabling agricultural robots to handle diverse geometries and disturbances. To address the limitations of GPS- and IMU-based navigation in high-density agricultural environments, the second study was carried out to investigate the feasibility of using a modular sensing system with distributed data processing that benefits from multiple infrared and Time-of-Flight sensors integrated with CANBUS-based communication. These sensors provided data for a fuzzy knowledge-based controller to enable the robot with reactive collision avoidance capabilities. Simulation and field tests validated both the proposed control algorithm and the system's performance under outdoor conditions. Simulation results indicated that the controller maintained the minimum allowed distance between the left and right sides of the robot and obstacles for each track: 0.4 m for the oval-shaped track, and 0.1 m for the 90-degree curved track. The control output for both experiments was proportional to the steering angle of the real robot and varied between −0.22 and 0.22, corresponding to −22° and 22°. During the field experiments, the average distance between the robot and the left bushes, as measured by the three sets of infrared sensors mounted on the left side of the robot, were 17.5, 18.5, and 17.1 cm. These values for the right bushes were found to be 13.5, 18.9, and 16.6 cm. From the perspective of exchanging sensor data with the controller, the CANBUS network was found to be robust and reliable, with 100% of the data transmitted during the experiment without any interruption. The modular hardware design facilitated scalability and maintenance, while the distributed data processing approach reduced computational load. The proposed solution offers a cost-effective alternative to high-end LiDAR-based sensing, ensuring accessibility for diverse agricultural applications. This study suggested that the performance of the collision avoidance system could be significantly improved in future studies by implementing a supervised machine learning algorithm. Given the challenges and limitations of achieving full autonomy in unstructured agricultural environments, where autonomous navigation is difficult and requires heightened safety measures, the third study focused on addressing the role of human operators in pathfinding and supervision through Teleoperation. Earlier studies have shown that local networks with long-range wireless communication, combined with digital twin concepts, are promising solutions for robot teleoperation. The third study aimed to explore the feasibility of this approach by remotely sending and receiving commands to a mobile robot in berry orchards from distances ranging between 300 and 3000 meters. The primary goal was to assist the robot in managing complex navigation tasks, such as turning at the end of rows, using the digital shadow of the robot that mirrored the robot’s state. This allowed the operator to provide informed guidance in challenging scenarios, such as row-end turns and obstacle avoidance. Field experiments demonstrated the feasibility of this approach, with an average packet loss of 12% at distances up to 2300 m. Despite challenges with latency and stability, the system proved capable of transmitting essential data for navigation, maintaining a robust communication link, and providing real-time situational awareness. Optimizations in simulation and message handling addressed computational and calibration issues, highlighting the potential of integrating 5G networks for further enhancements in speed, reliability, and visual feedback. This approach offers a scalable solution for extending the operational range and safety of agricultural robots in dynamic environments.