Obstacle-aware Adaptive Informative Path Planning for UAV-based Target Search

Abstract

The autonomous target search problem for Unmanned Aerial Vehicles (UAV) in urban environments requires solving a 3D path planning problem for maximal information gain, given a restricted flight duration. In this paper, we propose a general, Obstacle-aware Adaptive Informative Path Planning (OA-IPP) algorithm for the target search problem which uses active perception. The main contribution is the layered optimization approach that balances the exploration-exploitation trade-off through a Bayesian Optimization (BO) framework and simultaneously optimizes the 3D path using a standard optimizer. The planner simultaneously trades off between information gain, field coverage, altitude-dependent sensor performance, collision avoidance, target re-observation and Field of View (FoV) while planning. Through experiments in a simulated environment, we show that the proposed approach outperforms a pure exploratory IPP planner, a coverage planner, and a random sampling planner by demonstrating the fastest decrease in error related to target position estimates. Furthermore, we demonstrate the planner in simulations of varying complexity and obstacle density, demonstrating its applicability to a range of environments. Finally, we combine the proposed planning approach with an existing human detection pipeline, and demonstrate its efficacy in locating human victims in a realistic simulated environment.