Stealthy behavior simulations based on cognitive data

Predicting stealthy behaviors plays an important role in game design. It is, however, difficult to automate this task because interaction between human and dynamic environments is not easy to compute and simulate. In this note, we present a reinforcement learning method for simulating stealthy movements in dynamic environments. We use an integrated method of Q-Learning and Artificial Neural Networks (ANN) to implement an action classifier. Experimental results showed that our simulation agent responds sensitively to dynamic situations and thus can be helpful for game level designers to determine various game factors.