Learning-based Hierarchical Framework for Strategic and Tactical Planning

The emergence of the Advanced Air Mobility (AAM) paradigm necessitates the development of increasingly autonomous air traffic management solutions to handle the projected scale and density of operations. This paper introduces a novel hierarchical conflict resolution framework that addresses the high-density and dynamic operations associated with the future of air mobility. Existing methods often focus on providing solutions to specific aspects of Conflict Detection and Resolution (CDR), such as only tactical or strategic deconfliction, which can lead to conflicts among the various components of a CDR system. Our proposed framework is specifically designed to be modular, allowing for distributed responsibility and scalability, while integrating strategic and tactical planning within a unified Conflict Resolution Module (CRM) that eliminates the potential for conflicting advisories. To address the challenges of a dynamic environment, the framework employs a hybrid approach combining tree-based search with learning-based methods. A Single-Player Variable-Step Monte Carlo Tree Search (MCTS) algorithm is utilized for strategic planning, offering a long-horizon perspective and enabling proactive prediction of potential conflicts such as demand-capacity imbalances and corridor congestion. For tactical planning, two controllers are proposed: a rule-based deterministic resolver and a multi-agent reinforcement learning (MARL) policy, showcasing the modularity of the framework. A Deconfliction Monitoring Module (DMM) continuously forecasts trajectories and routes conflicts to the appropriate planning level based on time-to-conflict thresholds. The framework is evaluated on three scenarios that highlight the coordinated action of the strategic and tactical planner to achieve deconfliction.