Noise-Aware UAS Flight Path Planning in Urban Environments with Reinforcement Learning

This thesis presents a comprehensive approach to mitigating noise pollution from Unmanned Aerial Systems (UAS) in urban environments through flight path planning using reinforcement learning (RL). The study focuses on Turin, Italy, leveraging its diverse urban architecture to develop a comprehensive model. A detailed 3D occupancy grid map, based on OpenStreetMap data, was created to represent buildings' locations and heights while a population density map was developed to account for demographic variances. The research develops a dynamic noise source model that adjusts noise emission levels based on UAV velocity, ensuring realistic noise impact predictions. Acoustic ray tracing techniques are utilized to simulate noise propagation, accounting for atmospheric absorption and reflections from urban structures, providing a detailed analysis of noise distribution. The core of this work is the application of the Deep Deterministic Policy Gradient (DDPG) algorithm within the RL framework. The algorithm is tailored to optimize flight paths by minimizing noise impact while balancing other factors like travel distance and energy efficiency. The RL agent learns to navigate complex urban landscapes, integrating penalties for idling, excessive distance, and abrupt maneuvers to refine its path planning strategy. Key findings reveal that the RL-based approach effectively reduces noise impact in urban settings, making it a viable solution for integrating UAVs into urban air mobility systems. The methodology is scalable and adaptable, with potential applications in various urban environments globally. This research contributes to the development of sustainable urban air mobility by addressing the critical issue of noise pollution, enhancing public acceptance and regulatory compliance.