Particle-based State Estimation and Communication for Game Agents

Artificial intelligence for games is of great interest as video games are more and more popular. Nowadays, people are looking for more realistic game agents that take actions in the same way as intelligent human beings. To estimate the next state of a target, a game agent observes its surrounding and learns from these observations. Several agents can communicate to share information in order to find the target efficiently. Particle filtering has been widely applied in game AI, because it is an efficient, nonparametric method to learn from observations and is effective even for nonlinear, non-Gaussian state space distribution. In this project, I apply particle filters on state estimation for each individual game agent and on the communication in multiagent settings. This report shows a few examples that I have tried and the results on how well two game agents can communicate via particle filters.