Qualitative Physics in Angry Birds

Abstract

In this paper, we present a program designed to successfully and autonomously play Angry Birds, which attempts to embrace motives of human players in their choices of targets they want to shoot at in a game play. The program comprises two modules: the representation module and the reasoning module. In the former, we introduce qualitative space representation that utilizes notions such as “to lie on,” “to lie to the right,” “to be a shelter of a target,” etc. The latter investigates how particular blocks of a structure behave once one of them has been hit. It includes two algorithms, namely vertical impact and horizontal impact. The first one is a novel method of investigating the behavior of complex structures after one of their constituent blocks gets hit. Namely, it predicts which elements of a structure fall if a supporting block gets destroyed. Horizontal impact, on the other hand, simulates force propagation between adjacent elements after one of them gets struck. We also describe experimental tests we have conducted in which Vertical Impact correctly predicted which blocks will fall in over 98% of investigated cases.