Master thesis: Machine learning for pilot decision support
Research Area
Fighter pilots operate in environments where timely decisions have to be taken based on uncertain information and where an erratic decision might have fatal consequences. The complexity of air combat operations is constantly increasing and the pilots will ultimately have to face high-capable opponents with more resources. In these situations, superior situational awareness and intelligent usage of the system are necessary to accomplish the mission. Furthermore, it is expected that the pilots will have to operate alongside unmanned aerial vehicles (UAV). The ability to act in a coordinated way with UAVs is expected to pose a high advantage.
The goal of the project
Investigate artificial intelligence (AI) and machine learning (ML) methods that can potentially improve pilots' decision-making capabilities in an air combat environment. The main research area involves using the Reinforcement learning (RL) approach to estimate the risk that pilots are subjected to and help to make decisions based on this information.
Workflow of the thesis
-
Get familiar with the air combat environment, aircraft dynamics, and reinforcement learning
-
Teach an agent how to escape an incoming missile attack in 3D simulation environment
-
Propose a potential solution to help the pilot in decision making
-
Verify your proposed solution by running a 3D simulation
-
(If time allows) Find a potential solution to unmanned/manned aerial vehicle collaboration based on the description above
Work environment
-
Ubuntu operating system
-
python programming language (including packages such as PyTorch)
-
jsbsim (aircraft flight dynamics models)
-
FlightGear (visualize aircraft)
Requirements
This project is done in collaboration with Saab Aeronautics, therefore, the student needs to be a Swedish citizen
Supervisors
Edvards Scukins, Ph.D. student, Division of Robotics, Perception, and Learning
email:
Petter Ögren, Professor, Division of Robotics, Perception, and Learning
email:
