Adaptive sampling for tracking in pursuit-evasion games

In this paper, we investigate target tracking with adaptive sampling in order to optimize the use of expensive and limited resources that Autonomous Vehicles (AVs) have at disposition in pursuit-evasion games. An adaptive sampling policy is developed in order to minimize energy consumption while satisfying performance guarantees such as, increased probability of detection over time, and maintenance of the targets in sensors' Field Of View (FOV). The approach is applicable to networks that have perfect knowledge of the workspace, but little or no prior information about the targets. Furthermore, we propose a predictor-corrector tracking filter that uses geometrical properties of targets' tracks to estimate their positions using imperfect and intermittent measurements. It is shown that this filter requires substantially less prior knowledge about the targets and measurement noise, and processing power than Unscented Kalman Filter (UKF) and Sampling Importance Resampling Particle Filter (SIR PF) while providing comparable estimation performance in scenarios with intermittent information. The proposed approach is validated both in numerical simulations and experiments involving heterogeneous ground and aerial vehicles.