Finite Receding-Horizon Incremental-Sampling Tree With Application to a Fixed-Wing UAV

Abstract

An algorithm called Finite Receding-Horizon Incremental-Sampling Tree (RH-IST) will be presented. It contains elements from the well-known Rapidly Exploring Random Tree algorithm and receding-horizon framework to provide a computationally inexpensive planning algorithm for vehicles with limited sensing capability in uncharted environments. To resemble a finite planning horizon property in the algorithm, a deterministic configuration space sampling is conducted on the surface of a sphere around the root of the tree. To produce near optimal trajectories, samples are connected according to their cumulative costs, resulting in an optimization heuristic. In the tree construction phase, each branch of the tree consists of the trajectory of a fixed-wing unmanned aerial vehicle. The steering method necessary to construct the branches comprises of a closed-loop time-domain simulation on guidance level with a non-linear 3-D mass point representation of the plant, a path following controller and a reference path based on Bézier-splines. The feasible best branch of the RH-IST is chosen to provide the input for the execution of the plant based on the evaluation of a cost-functional. Simulations demonstrate the adaptive behavior of the RH-IST in a 3-D urban environment with static and unknown obstacles.