In traditional off-road autonomous navigation systems, the predominant focus has been on utilizing geometric data, primarily from Light Detection and Ranging (LiDAR) sensors, in order to autonomously navigate through challenging terrains. However, integrating semantic vision data alongside geometric information can significantly enhance the understanding of the environment. This can be accomplished by fusing the geometric point cloud and semantic image data to create a comprehensive representation of the surroundings that include both the physical layout and the semantic meaning of terrains and objects.

A possible approach to achieve this is by creating a Semantic OctoMap, which is a 3D map that captures both the geometrics and semantics of the vehicle’s surroundings. With the help of the information stored in this Semantic OctoMap, various types of cost maps can be generated to prioritize different criteria for path planning, such as travel distance, navigation time and navigation risk. The navigation system developed in this project can effectively identify various types of terrains and obstacles, using this information to plan its path accordingly. This capability has proven to be a key advantage of the developed navigation system. On the other hand, the traditional navigation system evaluated, which primarily relies on geometric or LiDAR data, demonstrated challenges in accurately identifying and navi-gating through hazardous obstacles and terrains.

Furthermore, the criteria-based navigation, which prioritizes factors such as travel dis-tance, navigation time, and navigation risk, was proven to be efficient in guiding the vehicle through the diverse environment. The capabilities of this developed system make it suitable for use in applications ranging from military operations to search and rescue missions, infra-structure inspection, environmental monitoring, precision agriculture and others.