For the reinforcement learning of agents in virtual environment, a flight simulation model architecture is designed by combining the integration method, flight algorithm, external force calculation, earth model and other factors in flight simulation. This architecture realizes the separation of data and algorithm in the process of flight computation at the software level, and adopts inheritance and pointer registration mechanism to achieve unified and hierarchical management of data related to flight computation. The state transfer model of aircraft is designed to realize the smooth connection of flight states between different stages, which can realize efficient large-scale parallel computation in reinforcement learning. The architecture proposed in this paper provides a flexible and efficient interaction mechanism for enhancing reinforcement learning of aircraft agents.
A Flight Simulation Model Architecture for Reinforcement Learning
Lect. Notes Electrical Eng.
International Conference on Genetic and Evolutionary Computing ; 2021 ; Jilin City, China October 21, 2021 - October 23, 2021
2022-01-04
11 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch