Reinforcement learning based multi-perspective motion planning of manned electric vertical take-off and landing vehicle in urban environment with wind fields

Songyang Liu, Weizi Li, Haochen Li, Shuai Li

Abstract: Electric vertical-takeoff and landing (eVTOL) aircraft, known for their maneuverability and flexibility, offer a promising alternative to traditional transportation systems. However, these aircraft face significant challenges from various perspectives, including the need to increase energy efficiency, enhance passenger experience, and mitigate noise impact on urban environments. While mathematical modeling-based approaches have been employed for flight motion planning, they often struggle to adapt to dynamic and complex environments. In this work, we introduce a three-dimensional motion planning method based on deep reinforcement learning (DRL), tailored for manned eVTOL flights through urban wind fields. Our approach considers three crucial aspects: aircraft energy consumption, passenger experience, and noise impact on urban environment. We modify the Proximal Policy Optimization (PPO) algorithm and design comprehensive reward function that considers these objectives. By incorporating energy efficiency, passenger experience, and noise impact into our reward function, our method demonstrates improved policy learning compared to existing approaches. Comparative experiments conducted under various wind conditions show that our method outperforms commonly used techniques, effectively optimizing multiple objectives in challenging urban environments. Code of our work are available at https://github.com/cgchrfchscyrh/eVTOL_RL/tree/main.