Artificial Intelligence enabled crowd simulation for immersive virtual environments.

Abstract

Virtual agents, controlled with artificial neural networks, start to appear more frequently in 3D virtual environments. They bring additional layer of immersion into the scene. Recent innovations in the field became an inspiration for this research. Purpose of this work is creation of agent-oriented virtual crowd simulation, that will increase immersion of any given target scene. Crowd have to act natural and appear spontaneous. To complete described task Unity 3D game engine will be used with additional ML-Agents package. First part of the present text introduces research problem and some theoretical context. Subsequently development works are described, supplemented with explanations and justifications. Then, training of neural networks is analysed and performance assessed. Final part of the works features migration of developed assets onto the provided target scene. At the end conclusions will be drawn and future development course set. During the work, in the stages mentioned above, focus was set on versatility of programmed assets, performance challenges and quality of developed crowd behavior.