Aircraft Simulation in Game Development Using Reinforcement Learning
Md. Navid Redwan Talha, Tonmoy Hassan, Al Sabab, Shaila Sharmin Mithila, Syed Hasib Ahmed · 2025
The progress in technology has provided opportunities for innovative solutions to intricate challenges. One possible method is employing reinforcement learning to model flying trajectories in intricate environments. Game development is a discipline that involves intricate reasoning and dynamic interplay between the user and the game environment. By employing several gaming engines, developers are now able to replicate real-life situations through the implementation of diverse machine learning methods. Aircraft simulation in game creation using reinforcement learning involves creating a visual depiction of real-life settings where aircraft may navigate complex environments without direct input from a human user. Currently, reinforcement learning is not widely applied in game development, particularly in simulation-based path finding techniques. This algorithm approaches possess the efficacy and capacity to generate sophisticated neural networks capable of directing an agent to do certain tasks. The aim of this project is to create aircraft simulations for game development by utilizing reinforcement-learning techniques, so that it can provide a foundational idea of the usage of this algorithm in path-detection based decision-making techniques. The goal is to demonstrate the effectiveness of reinforcement learning in a real-world scenario, where the aircraft independently assesses and selects its flying trajectory. The system will undergo testing in three distinct phases, involving the utilization of Blender3D, Unity 3D, and Anaconda prompts. The results will then be compared using TensorFlow. Several training sessions will be conducted in various environments using the Anaconda environment to optimize the outcomes. In the latter stages of development, a dynamic user interface will be implemented to enhance the user's experience. The method is anticipated to produce 152% improved AI-trained data, which can be utilized for constructing extensive simulation and game-projects.