Deep Learning-Based Complete Coverage Path Planning With Re-Joint and Obstacle Fusion Paradigm

With the introduction of autonomy into the precision agriculture process, environmental exploration, disaster response, and other fields, one of the global demands is to navigate autonomous vehicles to completely cover entire unknown environments. In the previous complete coverage path planning (CCPP) research, however, autonomous vehicles need to consider mapping, obstacle avoidance, and route planning simultaneously during operating in the workspace, which results in an extremely complicated and computationally expensive navigation system. In this study, a new framework is developed in light of a hierarchical manner with the obtained environmental information and gradually solving navigation problems layer by layer, consisting of environmental mapping, path generation, CCPP, and dynamic obstacle avoidance. The first layer based on satellite images utilizes a deep learning method to generate the CCPP trajectory through the position of the autonomous vehicle. In the second layer, an obstacle fusion paradigm in the map is developed based on the unmanned aerial vehicle (UAV) onboard sensors. A nature-inspired algorithm is adopted for obstacle avoidance and CCPP re-joint. Equipped with the onboard LIDAR equipment, autonomous vehicles, in the third layer, dynamically avoid moving obstacles. Simulated experiments validate the effectiveness and robustness of the proposed framework.