Topological path planning for autonomous information gathering

Seth McCammon, Geoffrey A. Hollinger

Abstract: In this paper, we present two novel algorithms for information space topological planning that identify topological features in an information field and use them to plan maximally informative paths for a robot in an information gathering task. These features provide a way to rapidly incorporate global context into the informative path planning process by partitioning the state space or the path space of a robot. Our first algorithm, hierarchical hotspot information gathering, uses a topological state space partitioning by constructing a high-level map of information hotspots. We then solve a global scheduling problem over the topological graph, the solution of which is then used for path planning by a set of local greedy coverage planners within each hotspot. Our second algorithm, Topology-Aware Self Organizing Maps, extends the Self Organizing Map algorithm to discover prominent topological features in the information function. These features are used to perform a topological path space decomposition to provide a Stochastic Gradient Ascent optimization algorithm with topologically diverse initialization, improving its performance. In simulated trials and field experiments, we compare the tradeoffs of these two approaches and show that our methods that leverage topological features of the information field consistently perform competitively or better than methods that do not exploit these features, while requiring less computation time.