Robust planning for autonomous parafoil

Abstract

Parafoil trajectory planning systems must be able to accurately guide the highly non-linear, under-actuated parafoil system from the drop zone to the pre-determined impact point. Parafoil planning systems are required to navigate highly complex terrain scenarios, particularly in the presence of an uncertain and potentially highly dynamic wind environment. This thesis develops a novel planning approach to parafoil terminal guidance. Building on the chance-constrained rapidly exploring random tree (CC-RRT) [1] algorithm, this planner, CC-RRT with Analytic Sampling, considers the non-linear dynamics, as well as the under-actuated control authority of the parafoil by construction. Additionally, CC-RRT with Analytic Sampling addresses two important limitations to state-of-the-art parafoil trajectory planners: (1) implicit or explicit constraints on starting altitude of the terminal guidance phase, and (2) a reactive or limitedly-proactive approach to handling the eect of wind uncertainty. This thesis proposes a novel formulation for the cost-to-go function, utilizing an approximation of the reachability set for the parafoil to account for the eect of vehicle heading on potential future states. This cost-to-go function allows for accurate consideration of partially planned paths, effectively removing strict constraints on starting altitude of the terminal guidance phase. The reachability set cost-to-go function demonstrates considerably improved performance over a simple LQR cost function, as well as cost-to-go functions with a glide-slope cone bias, demonstrating the eectiveness of utilizing the reachability set approximation as a means for incorporating heading dynamics. Furthermore, this thesis develops a multi-class model for characterizing the uncertain effect of wind. The wind model performs an online classication based on the observed wind measurements in order to determine the appropriate level of planner conservatism. Coupling this wind model with the method for sampling the analytic uncertainty distribution presented in this thesis, the CCRRT with Analytic Sampling planner is able to eciently account for the future eect of wind uncertainty and adjust trajectory plans accordingly, allowing the planner to operate in arbitrary terrain configurations without issue. CC-RRT with Analytic Sampling performs exceptionally well in complex terrain scenarios. Simulation results demonstrate signicant improvement on complex terrain relative to the state-of-the-art Band-Limited Guidance (BLG) [2], drastically reducing the worst case and average target miss distances. Simulation results demonstrate the CC-RRT with Analytic Sampling algorithm remains un-affected as terrain complexity increases, making it an ideal choice for applications where difficult terrain is an issue, as well as missions with targets with drastically dierent terrain conditions. Moreover, CC-RRT with Analytic Sampling is capable of starting terminal guidance at significantly higher altitudes than conventional approaches, while demonstrating no signicant change in performance.