Abstract
Cooperative Artificial Intelligence (AI) in Human-Agent interaction is a complex but interesting subject that has not yet reached the necessary behavior to be considered satisfactory. In this document, we present how we achieved our goal of creating an artificial players capable of effectively solving Geometry Friends levels, using a promising previous approach, that uses the Rapidly-Exploring Random Trees (RRT). Starting on this solution, we developed our agents by adding new strategies to the algorithm, replanning, and a new controller. When our agents proved capable of solving single-player levels a step was taken towards an initial approach for cooperation with human players. Our final solution is a pair of agents that can not only solve most of the presented challenges of single-player levels, with some current limitations, but also play with human players in simple cooperation levels. After a user testing session, the most human players affirmed the experience was positive and that our agents presented capabilities of cooperation and adaptation.