Robustness for large scale UAV autonomous operations

Unmanned aerial vehicles (UAVs) need multiple operators as of today. Because of limitations of human attention and sensory cueing the operators cannot use all of the data coming in through these sensors. Large scale autonomy would mean the following: fewer operators, larger areas of coverage, and more vehicles. Rigorous optimization of decision theory based approaches to handling this problem suffer from speed limits to real-time computations. Soft-computing methods based on heuristics (AI, reasoning, neural networks and fuzzy logic) are able to handle certain circumstances but cannot supply any guarantees of performance in unstructured environments. We provide solutions based on rigorous approaches, but avoid heavy real-time computation through off line processing. For mapped regions, we convert maps into traversability graphs using trapezoidal map/voronoi type algorithms. We then find the shortest permissible paths for different vehicles using Dijkstra's algorithm and then preening the allowable paths using constraints on individual vehicles. Out third step consists in determining the minimum time taken by a given vehicle over those paths. We insert margins of safety at each level of this hierarchy: buffer zones (size of buffer zone is a ten meters extended polygon around each building) around mapped structures to account for map errors; buffer zones around flight paths to account for position uncertainty of vehicles; performing 1-D optimal control within limits of the vehicles performance so that vehicles can slow down or speed up in response to unexpected events.