Advancing Autonomous UAV Target Localization in GPS-Denied Environments

This research introduces a novel approach to enhance the robustness of autonomous UAV target localization in GPS-denied environments. Typically, localization in such environments relies on utilizing received signal strength (RSS) transmitted by the target and employing the Q-learning algorithm to determine agent actions. Usually, there are two approaches to defining Q-learning's state, both using only a single variable as the Q-learning's state. One of those approaches is potentially hard to converge, while the other is susceptible to unrealistic assumptions in a real-world deployment. This study proposes an improved Q-learning algorithm by incorporating multiple variables as states, addressing the limitations of existing approaches that use a single variable. In this research, it will be proven that if the assumptions don't hold, the performance of the proposed method by another research will be messed up. In addition, it's proved that a new approach is proposed in this research able to outperform the approach using only single-variable as Q-learning state in terms of speeds of convergence and robustness, even when the assumptions are not met.