Procedural Content Generation Using Reinforcement Learning for Disaster Evacuation Training in a Virtual 3D Environment

This research addresses the need for effective disaster evacuation training methods by proposing a virtual reality system that utilizes Reinforcement Learning Procedural Content Generation (RL-PCG) algorithms. The aim of this study is to provide a cost-effective and safe way to conduct disaster evacuation preparedness training, surpassing the limitations of traditional real-life drills. The paper’s objectives encompass the design of a novel 3-layer PCG architecture for generating realistic disaster simulations in virtual reality, the implementation of a working prototype for fire disaster scenarios, and the evaluation of the proposed system’s effectiveness through comparison with existing RL agents. Significant findings include the superiority of the RL-PCG agent in generating diverse and realistic disaster scenarios with faster training time and lesser number of steps, even with limited processor capabilities. In conclusion, this research establishes that the RL-PCG Scenario for Disaster Evacuation Training in VR is a more effective method, leading to improved disaster preparedness for individuals, and opens avenues for further advancements in disaster training using virtual reality and reinforcement learning technologies. For a video demo of this work, please visit
https://youtu.be/3WZnQOfUP94
.