Vision-based Exploration Algorithms for Rough Terrain Modeling Using Triangular Mesh Maps

The purpose of this paper is to develop a new exploratory approach based on a triangular mesh map for automatic modeling of a large rough agricultural environment. A triangular mesh map was used to represent the agricultural field surface because of its ability to represent large rough areas efficiently. A terrain map is built incrementally during exploration, using 3D image sensor readings. A 3D image sensor model, with attributes similar to a camera or laser sensor, was used in the simulation. A two-stage exploring policy was used to plan the next-best view by considering both the distance and elevation change in the cost function. In the first stage of exploration, the robot travels to the outer boundary between the explored and unexplored terrain, while in the second stage it fills in the hole left by the first stage. Previous work considered distance as the only traveling cost. In this work, the slope factor is also included in the cost function because the mobile robot needs more energy to overcome the changes in elevation. A line sweeping approach based on the bug concept is also presented to identify a path for complete coverage of the terrain. The two methods are implemented and validated in simulation. A complete comparison of the traveling distance, time consumption, and number of scans recorded using the two methods is presented to show the effectiveness of the two-stage exploration algorithm.