Cross-Modal Knowledge Distillation for autonomous urban air mobility routing during mission-critical recovery windows Introduction: A Learning Journey into the Skies It was a rainy Tuesday evening when I first stumbled upon the problem that would consume my next six months. I was debugging a reinforcement learning agent for drone delivery routing—a relatively straightforward task of navigating a fleet of eVTOLs (electric Vertical Take-Off and Landing aircraft) across a simulated city. The agent performed admirably under normal conditions, achieving 94% on-time delivery rates. But when I simulated a sudden storm—a "mission-critical recovery window" where we had only 15 minutes to reroute 47 aircraft after a major hub failure—the agent collapsed. It froze, outputting nonsensical paths that violated no-fly zones and ignored battery constraints. That failure sparked my deep dive into cross-modal knowledge distillation.…