A Methodology for Learning Airplane Descent Patterns

In this study, we trained a plane to land without human assistance.

In this study, we trained a plane to land without human assistance. We did this using reinforcement learning, a type of machine learning where the agent (the AI “pilot” in this case) runs iterations of the task and adjusts its methodology depending on this interaction with its environment. This agent-environment interaction is called the Markov Decision Process, and is the heart of reinforcement learning. This is an important problem to solve as reinforcement learning applications like these can be used to improve the safety and quality of life for people who are regularly traveling. Additionally, the fact that reinforcement learning can be used to solve problems like these implies further applications in complex fields like flight and other modes of transportation. To figure out how to train a plane to land, we used OpenAI Gym to build a simulated environment where we tested different landing maneuvers. Initially, we experimented with optimizing the angle of attack for variables like time taken to land and descent rate, but ultimately found it most effective to optimize for having the lowest velocity when the plane reaches the ground. Further research could potentially experiment with how different variables such as speed, acceleration, and angles of the left and right ailerons affect the plane’s landing.

Explore More!

Research Paper