Learning-Based iLQR for Multi-UAV Collision-Free Planning in 3D Environments

Flying along collision-free trajectories is crucial for multiple unmanned aerial vehicles (UAVs) to ensure efficient and safe autonomous operations in shared airspace. This paper aims to address the collision-free planning problem for multiple UAVs in three-dimensional (3D) space using deep learning and iterative linear quadratic regulator (iLQR) techniques. First, the long short-term memory (LSTM) network is employed to learn local system models around UAV trajectories. By incorporating LSTM into iLQR, the UAVs can be controlled without the need for intricate dynamics modeling. This approach takes obstacle limitations into account to prevent collisions. Second, a regularization method is used to enhance the robustness of numerical calculations in LSTM-iLQR. Third, the interaction among UAVs is formulated as a differential game. This game is then transformed into an optimal control problem and solved using iLQR to avoid inter-UAV collisions. Finally, the effectiveness of the proposed method is verified through simulations.