Flight route planning
||Flight route planning|
||Arne Jönsson <firstname.lastname@example.org>|
||2021-04-15 – 2021-11-01|
This project is a candidate thesis on behalf of SAAB which aims to explore how well various classical and reinforcement learning algorithms can perform flight route planning and flight control in environments with natural and artificial threats. Here, the focus is placed primarily on how the addition of memory mechanisms influences the ability to plan. Numerous simulated aerodynamic physics worlds are run concurrently on the processor, while the neural network inference and training are done using a CUDA-device.
The research questions to be answered are the following:
- How well can the ML planning models rival a hardcoded approaches?
- How well does different planning algorithms function depending on search space size in terms of execution time, memory usage, and plan quality?
- For the models that offer insight into the planning process, what strategies can be observed as having been developed?