Sampling-Based Exploration Strategies for Mobile Robot Autonomy

A novel, sampling-based exploration strategy is introduced for Unmanned Ground Vehicles (UGV) to efficiently map large GPS-deprived underground environments. It is compared to state-of-the-art approaches and performs on a similar level, while it is not designed for a specific robot or sensor configuration like the other approaches. The introduced exploration strategy, which is called Random-Sampling-Based Next-Best View Exploration (RNE), uses a Rapidly-exploring Random Graph (RRG) to find possible view points in an area around the robot. They are compared with a computation-efficient Sparse Ray Polling (SRP) in a voxel grid to find the next-best view for the exploration. Each node in the exploration graph built with RRG is evaluated regarding the ability of the UGV to traverse it, which is derived from an occupancy grid map. It is also used to create a topology-based graph where nodes are placed centrally to reduce the risk of collisions and increase the amount of observable space. Nodes that fall outside the local exploration area are stored in a global graph and are connected with a Traveling Salesman Problem solver to explore them later