A Survey of Current Applications of Inverse Reinforcement Learning in Aviation and Future Outlooks

Many problems in aviation can be characterized as sequential decision-making problems under uncertainty, such as air traffic management and flight delay prediction. One approach to solving such problems is to learn from expert demonstrations. Inverse reinforcement learning (IRL) is one such class of methods, where a reward function is learned given expert demonstrations. However, while IRL has shown promising results in recent years, for example in autonomous vehicle path planning, we identify a significant, quantifiable gap in applications of IRL in aviation relative to other domains. In this paper, we formally introduce IRL, provide a comprehensive summary of foundational methods, review how IRL has been applied in the field thus far, discuss challenges that may explain the IRL gap, and explore potential future applications of IRL for aviation.