UAV Path Planning for Wildfire Tracking Using Partially Observable Markov Decision Process

Real-time monitoring wildfire spread is of utmost importance to the decision making of the wildfire management. Unmanned aerial vehicles (UAVs) can provide a mobile sensing platform to perform this challenging task. In this paper, a path planning algorithm is designed to enable a group of UAVs to autonomously track wildfire fronts that develop in random directions. The wildfire evolution is generated from a high-fidelity simulation model. Since the full state of the wildfire is unknown, the path planning is formulated as the partially observable Markov decision process (POMDP), which is a dynamic optimization problem and solved approximately by the computationally efficient nominal belief state optimization method. The approach can plan the path autonomously, and the UAVs are able to track any randomly spreading fire fronts through a simple linear model. In addition, the challenges such as collision avoidance and practical constraints of the control variables can be taken into account in the framework of POMDP. Comprehensive simulation results demonstrate the effectiveness of the proposed path planning algorithm.