Rapidly-Exploring Random Tree approach for Geometry Friends

Abstract

Geometry Friends (GF) is a physics-based platform game, where players control one of two characters (a circle and a rectangle) through a series of both individual and cooperative levels. Each level is solved by retrieving a set of collectibles. This paper proposes an approach using Rapidly-exploring Random Trees (RRTs) to find a solution for the individual levels of Geometry Friends. Solving a level of GF is divided into two subtasks: (1) planning the level; and (2) executing in real time the sequence of moves required to fulfill the plan. We use our RRT approach to solve (1) and a Proportional Integral Derivative (PID) controller to guide (2). The quality of the agent implemented was measured in the 2015 GF Game AI Competition. Results show that our agents are able to plan both public and private levels and are able to control their motion in order to finish most of them.