Self-Calibrating UAV Navigation: Reinforcement Learning Approaches for Horizontal Trajectory Estimation

Accurate unmanned aerial vehicle (UAV) trajectory estimation is essential for autonomous navigation, particularly in GPS-denied environments. Visualodometry and simultaneous localization and mapping (SLAM) approaches require precise camera intrinsic parameters, which are typically obtained through predefined or offline calibration. Instead, in this work, we propose a reinforcement learning (RL)-based self-calibration framework that estimates camera intrinsic parameters directly from monocular video sequences, without requiring prior knowledge of the camera, environment, or calibration targets. This intrinsic parameter estimation is then leveraged to achieve robust UAV trajectory estimation using only video data. We formulate the problem as a sequential decision-making task, where an RL agent iteratively refines intrinsic parameters (focal lengths and principal point) by minimizing projection error. The proposed framework is implemented using the Soft Actor-Critic (SAC) algorithm, which is well-suited for continuous action spaces and promotes efficient exploration. We validate our approach on real-world datasets, which provide ground truth intrinsic parameters and trajectory data. Our results demonstrate that the estimated intrinsic parameters enable effective UAV trajectory reconstruction in GPS-denied environments, showing promising results. Thereby, this approach enables the estimation of the UAV's 3D trajectory without prior knowledge of the camera.