Efficient Planning for Safe Air Traffic Control With STL Constraints Using Deep Reinforcement Learning

Air traffic control (ATC) is an important problem in aerial traffic management, which can be formulated as a planning problem with predefined signal temporal logic (STL) specifications. This formulation leads to a nonconvex mixed integer programming (MIP) problem which is computationally expensive due to the large amount of variables and constraints. However, the strict safety requirements and the time limitations of practical applications require solving feasible solutions efficiently. In this study, we enhance the efficiency of solving MIP by leveraging a large neighborhood search (LNS) algorithm, which aims to obtain near optimal solutions with reduced computational complexity by iteratively selecting subsets of variables and constraints to form smaller subproblems.. This subset is automatically chosen by a graph convolutional neural network (GCN) trained using a deep reinforcement learning (RL) algorithm with non-labeled data extracted from an off-the-shelf solver. We evaluate the proposed RL-based LNS approach on a dataset derived from a basic taxiing ATC scenario. Experimental results demonstrate that our method achieves a 68.6% reduction in computation time compared to the open-source heuristic solver SCIP, while maintaining high solution quality.