MIT researchers have developed an innovative method inspired by the movie “Top Gun: Maverick” to address complex challenges related to aircraft stability and evasion. Using a machine learning approach, their technique surpasses current safety standards and enhances stability by a factor of ten. The method successfully guided a virtual fighter jet through a narrow passage, impressing even experts in high-dimensional dynamics. Traditional approaches simplify complex stabilise-avoid problems using mathematical techniques, while reinforcement learning trains an agent through trial and error. However, balancing stability and obstacle avoidance in these problems is challenging. The MIT researchers tackled the problem in two steps: redefining it as a constrained optimization problem and transforming it into the epigraph form for deep reinforcement learning. To handle the epigraph form, the researchers derived new mathematical expressions specific to their system. They combined these with established engineering techniques, creating a controller that outperformed baselines by preventing crashes and achieving stable alignment with the desired goal. This technique has potential applications in designing controllers for dynamic robots and assisting in stabilising autonomous vehicles. It excels in extreme scenarios, providing reinforcement learning with safety and stability guarantees for mission-critical systems. Future enhancements will focus on better handling uncertainty during the optimization process and testing the algorithm on physical hardware. The researchers aim to bridge the gap between model dynamics and real-world dynamics for practical implementation.