Fire front path planning and tracking control of Uncrewed Aerial Vehicles using deep reinforcement learning

Abstract

This study develops a unified path planning and control framework based on reinforcement learning for Uncrewed Aerial Vehicles (UAVs) operating in dynamic wildfire environments. The Deep Deterministic Policy Gradient (DDPG) algorithm facilitates tracking fire evolution through a structured architecture comprising high-level planning and low-level control components. The path planner computes the linear velocity and refines the heading angle by incorporating the fire’s directional properties to generate the target trajectory. The low-level controller ensures stable trajectory tracking by adaptively tuning the control gains during the learning process. The closed-loop stability of the overall system is analytically validated using Lyapunov-based analysis. The framework is evaluated using the FARSITE fire area simulator, calibrated with real-world wildfire data. The simulation results demonstrate that the framework generates smooth planning variables, provides adaptive tracking, and remains robust against a range of external disturbances.