Funding entity Airbus Defence and Space S.A.U
This project aims to develop a learning-based trajectory planning system for UAVs operating in complex environments with multiple agents, dynamic obstacles, and static obstacles. Traditional optimization-based planners, while precise, are often computationally expensive and fragile in dynamic settings. To address these limitations, the proposed approach leverages imitation learning to initialize a reinforcement learning policy, combining the efficiency of learned behaviors with the adaptability of Reinforcement Learning. This hybrid method enables fast, robust, and scalable trajectory planning capable of handling uncertainty and complex UAV dynamics in real time.
Layman's summary: This project teaches drones to fly safely around dynamic and static obstacles, and other drones by learning from examples and practice. The goal is fast, smart decision-making in complex, changing environments.
Techniques employed: Reinforcement Learning, Trajectory Optimization, Splines
Airbus_AI_2
