Learning-Based Artificial Potential Field Path Planning for Agricultural UGVs

The agricultural sector's growing importance in ensuring food security and sustainability has driven the development of Agriculture 4.0, leveraging advanced technologies. The potential of Unmanned Ground Vehicles (UGVs) in automating agricultural tasks highlights the necessity of efficient motion planning solutions. Nevertheless, conventional point-to-point (P2P) path planning algorithms frequently do not allow the selection of a specific path planning strategy. This paper presents a novel Learning- Based Artificial Potential Field (L-APF) algorithm, integrating Machine Learning techniques with potential fields for local point-to-point (P2P) navigation. A simplified AlexNet Convolutional Neural Network (CNN) version is employed to enhance UGV guidance. The method is validated in Matlab through numerous simulations in different scenarios and with different required path planning strategies. Results show the L-APF's robustness and adaptability, allowing for effective navigation in dynamic agricultural environments. The conclusion highlights its potential for real-world agricultural applications, particularly in tasks requiring precise navigation, such as spraying and crop monitoring.