Predicting Opponent Actions by Observation

摘要

In competitive domains, the knowledge about the opponent can give players a clear advantage. This idea lead us in the past to propose an approach to acquire models of opponents, based only on the observation of their input-output behavior. If opponent outputs could be accessed directly, a model can be constructed by feeding a machine learning method with traces of the opponent. However, that is not the case in the Robocup domain. To overcome this problem, in this paper we present a three phases approach to model low-level behavior of individual opponent agents. First, we build a classifier to label opponent actions based on observation. Second, our agent observes an opponent and labels its actions using the previous classifier. From these observations, a model is constructed to predict the opponent actions. Finally, the agent uses the model to anticipate opponent reactions. In this paper, we have presented a proof-of-principle of our approach, termed OMBO (Opponent Modeling Based on Observation), so that a striker agent can anticipate a goalie. Results show that scores are significantly higher using the acquired opponent's model of actions.