A Comparison of Risk Sensitive Path Planning Methods for Aircraft Emergency Landing

Determining the best site to land a damaged aircraft presents some interesting challenges for standard path planning tech-niques. There are multiple possible locations to consider, the space is 3-dimensional with dynamics, the criteria for a good path is determined by overall risk rather than distance or time, and optimization really matters, since an improved path cor-responds to greater expected survival rate. We have investi-gated a number of different path planning methods for solving this problem, including cell decomposition, visibility graphs, probabilistic road maps (PRMs), and local search techniques. In their pure form, none of these techniques have proven to be entirely satisfactory – some are too slow or unpredictable, some produce highly non-optimal paths or do not find certain types of paths, and some do not cope well with the dynamic constraints when controllability is limited. In the end, we are converging towards a hybrid technique that involves seeding a roadmap with a layered visibility graph, using PRM to ex-tend that roadmap, and using local search to further optimize the resulting paths. We describe the techniques we have in-vestigated, report on our experiments with these techniques, and discuss when and why various techniques were unsatis-factory. 1.