Data-driven scenario modeling and generation for virtual training system

• Main Author: Yin, Haiyan
• Other Authors: Cai Wentong
• Format: Final Year Project
• Language: English
• Published: 2014
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/59085
• Institution: Nanyang Technological University

The rising penetration of virtual training has necessitated the development of meaningful training scenarios to ensure training efficiency. For mission-based virtual training, the scenarios often consist of two common aspects: On the one hand, scenarios are associated with trainer's desired mission objectives which specify the tasks or knowledge to be trained on; On the other hand, they are preferred to be customized to individual trainee for an increased training efficiency. However, developing such a scenario is costly, because initializing a scenario that fulfills the desire of both the trainer and the trainee is comprehensive and often requires a certain amount of manual effort. Moreover, scenarios are consumed fast because there often lacks replay-ability. This study aimed to design an automated scenario generation framework for virtual training. The framework involved the trainer's desired mission objectives to control the difficulty of scenario and the trainee's skill levels to realize customization. In addition, a data-driven approach was proposed to improve the evaluation of scenario's difficulty levels. In this approach, player's performance data on the simulation was collected and trained to construct artificial neural networks (ANNs). To facilitate the collection of player's performance data, we also modeled an AI-Player which can imitate the real human player's playing behaviors. We conducted an empirical study on an interactive game with food distribution mission to demonstrate the efficiency of the automated scenario generation framework for virtual training.