Overview: |
The developments of 3D First-Person Shooter (FPS) games will be present in the beginning of this session. It comes up with the challenge to design a proper NPC that helps to train and encourage the freshman, and keep the old player continually stimulated. The challenges include the basic ability of NPC to handle the full game, exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human. To develop the basic full-game capability, the state-of-art Reinforcement Learning (RL) algorithm is applied to train AI agents from scratch, following the timeline of the sequential tasks of the FPS game. In the meantime, Generative Adversarial Imitation Learning (GAIL) is applied to increase the behavioral similarity of the AI to human. Then, some demonstrations of these properties in a case study of Call of Duty Mobile (CODM) are given. Finally, the exploration of the potential commercial value of the intelligent NPC is illustrated in a successful case.
|