Occupancy Grid Map Completion Based on Geometric Features and Door Discovery

Occupancy grid maps are one of the most robust environment representations to use for problems in robotics such as exploration, navigation, etc. If the partial occupancy maps are used after predicting the uncompleted parts of the map, it is expected that the performance of these tasks will be increased. Therefore, in this paper, we suggest completing partial occupancy maps by using door discovery. In order to do this, initially we find door positions by utilizing geometrical features of the maps and then we implement map completion using image processing techniques based on these door positions. The experimental results show that our approaches are promising with average IoU score up to 0.8790 and have potential to improve the robotic tasks.