A Map Segmentation Method Based on Deep Learning for Complete Coverage Path Planning

Path planning is one of the key technologies for autonomous robot navigation and operation. Based on global and local planning, full-coverage path planning is an essential method for solving tasks such as cleaning, inspection, and harvesting in a given area. In this work, complete coverage path planning of a known area is studied. First, navigation and operation within simple graphics like rectangles and triangles are realized. Then, a map segmentation method based on image processing and deep learning is proposed to segment the map built through the lidar into several basic graphics. Full coverage path planning is then carried out for each basic graphic, and finally, the sub-areas are connected to achieve a complete coverage of the entire region. To validate the feasibility of the method, a coverage rate calculation method was proposed. A series of experiments were conducted in simulation worlds. The experimental results showed that the proposed method can help the robot to achieve full coverage driving and operation in complex areas. Combining with the robot's operating range, the final achieved operational coverage rate was 97% or more. In the presence of obstacles, a maximum coverage rate of 95 % was still achieved through a local obstacle avoidance strategy.