Optimizing Air-to-Air Missile Guidance using Reinforcement Learning

Optimizing Air-to-Air Missile Guidance using Reinforcement Learning

van Hoorn, Martijn (TU Delft Aerospace Engineering)

Hartjes, S. (mentor)

Delft University of Technology

Aerospace Engineering | Flight Performance and Propulsion

2019-03-26

To increase performance of air-to-air missile guidance, a novel guidance law is developed using reinforcement learning methods. This guidance law is based on behavior obtained from optimal control methods and subsequently aims to approximate its performance. The study compares the developed guidance law to a traditional guidance law and optimal solutions. It is established that the novel guidance law outperforms the implemented traditional guidance law in terms of range and time of flight to target. Generated trajectories mimic behavior of optimal control and a large fraction of optimal performance is achieved in terms of range (~90%) and time of flight (~95%).

Reinforcement Learning
Trajectory optimization
Guidance
machine learning
Missile
DDPG

http://resolver.tudelft.nl/uuid:6b26d24a-780c-47ee-9456-7af07490a317

2021-03-26

Student theses

master thesis

© 2019 Martijn van Hoorn