dc.description.abstract | Abstract: This thesis presents a sampling scheme for traversability mapping and a motion planning approach for terrain scenarios. The sampling scheme uses overlapping minimum enclosing cylinders to analyze a point cloud and to estimate terrain inclinations over a grid map. For motion planning, an objective cost function is defined, that allows penalization of an interval of inclination values based on varying a parameter 𝜆. RRT and RRT* are extended with a motion primitive scheme that ensures an implicit obstacle-clearance by performing collision checking in a minimum enclosing radius from every node in a tree. In the RRT-variant, accumulated costs through branches from the root node of the tree, in a field defined by the objective function, are saved at every node. A path is constructed by picking the node with the lowest accumulated cost from the root within a radius from the goal. The RRT*-variant also utilizes the cost field defined by the objective function, and employs a proposed BVP-rewiring scheme. For comparison, the motion primitive variants are modified to use inclination cost directly. Results show that the proposed motion primitive scheme produces more traversable paths when higher inclinations are penalized, indicating that the collision-checking approach is the significant contributor to higher path traversability for the sampling-based variants. This finding is supported by the results, as no apparent trend is observed with respect to 𝜆 when the motion primitive variants are compared directly, but a clear trend is observed with respect to the ideal benchmarks. Since no apparent trend is observed between the proposed algorithms, this suggests the RRT* optimization scheme does not produce more traversable paths compared to paths produced by the RRT-variants. The finding that collision-checking is the significant contributor to higher path traversability for the sampling-based variants, is also supported by the results from a simple non-holonomic version of FMM, which is included to show that for vehicles without netural-turn and reverse-motion capabilities, feasibility is not guaranteed by FMM alone. | en_US |