Reinforcement Learning for the active control of rockets

The focus of the PhD will be on the use of Reinforcement Learning for the active control of rockets for both terrestrial and extra-terrestrial applications. As has been demonstrated by SpaceX, the use of reusable rockets can significantly reduce the cost of launches and can fundamentally change the commercial landscape and what is possible in terms of frequency and payload to orbit.

Building on previous work at Bristol in the development of Reinforcement Learning (RL) for the flight control of drones, this PhD will develop RL based flight control systems for manoeuvring and hovering rockets. For challenging manoeuvres and operations in a diverse range of environments, these offer the potential for a robust and adaptable approach to the control of rocket propelled vehicles. Representative models will be used for developing baseline controllers, simulation for RL based training and experimental tests for validation. Extensions of the terrestrial models will allow for application to moon and mars-based challenges for reusable rockets.

Student: Jack Cunningham

University Partner: University of Bristol

Industry Partner: Airborne Engineering Ltd