Deep Reinforcement Learning for Optimal Takeoff Trajectory Design of an eVTOL Drone

The continuing development of electric vertical take-off and landing (eVTOL) aircraft presents promising opportunities to alleviate transportation congestion. However, designing and executing takeoff trajectories that optimally balance energy usage, passenger comfort, and aircraft constraints remains challenging. Conventional design optimization provides a solution for pre-defined flight conditions and environments, but is not ideal in real-world applications. In contrast, deep reinforcement learning (DRL) implements optimal policy making real-time decisions with no assumption of mathematical models. Seeing the lack of literature on DRL- based takeoff trajectory design of eVTOL aircraft, we implement and conduct DRL-based optimal takeoff trajectory designs for the Airbus ��³ Vahana drone in this work. Specifically, we aim to achieve optimal takeoff trajectory designs with the minimum energy consumptions while satisfying takeoff metrics (��.��., vertical displacement and horizontal velocity) by varying control profiles (��.��., power and wing angle to the vertical). Results revealed that the trained DRL policy managed to realize the takeoff by meeting all metrics. Moreover, the verification against the simulation-based optimization reference presented that the DRL-based optimal takeoff trajectory achieved 96.3% accuracy on the optimal energy consumption. The DRL-based optimal control profiles also showed similar trends as the optimal references.