Exit Decisions Inspired by Reinforcement Learning

Bounded rationality is one of the key components of human behaviors. Human action normally involves experience accumulation, exploration, and a combination of them. Though it is invisible and difficult to model, in this work, we propose a reinforcement learning-based model and applied it to a particular escaping case to evaluate the behavior of an intelligent agent who wants to optimize its efficiency to escape a typical map. The result resembles the tendency of humans in a procedure of escape including a strong tendency to move out near the exit, randomness when further away from the exit, and some tendency to move near the boundary of the map. Such behaviors are very close to human behaviors which indicates that the decisions made in emergence are mainly based on experience. Such a perception-action model of RL learning can be used in a multi-agent model to investigate collective behaviors.