Accepted Papers

Intuitive Rules Design Evaluation Methods and Case Study by Joseph Alexander Brown, Hamna Aslam, Munir Makhmutov and Giancarlo Succi

Brown et al. present an observational study approach for determining which rules of a board game can be intuitively understood from the game pieces themselves. The paper describes how to apply such a method, and presents some preliminary results from a case study and potential uses for the methodology.

Yang and Ontañón describe a method for using supervised learning algorithms to learn tree policies for Monte-Carlo Tree Search (MCTS) RTS game playing agents. They present results comparing the use of Bayesian classifiers and decision tree classifiers, and show that several of the models achieve promising results.

Bendekgey presents a novel representation for game traces in the card game Dominion. They compare the usefulness of their representation to other representations by clustering the game traces to find player strategies. They find that their representation is able to mitigate the effects of the game length on the clustering, and that interesting player strategy clusters emerge.

Extracting CCGs for Plan Recognition in RTS Games by Pavan Kantharaju, Santiago Ontañón and Christopher Geib

Kantharaju et al. present a greedy algorithm for automatically extracting Combinatorial Categorical Grammar (CCG) definitions from Real-Time Strategy (RTS) plan traces acquired game replays. Their results show that their approach is more scalable than current Hierarchical Task Network learning approaches.

Braylan and Miikkulainen describe clustering-based approaches for identifying and discretizing sections of game levels and maps based on environmental information, such as the fauna present in the sections. Their proposed approaches outperform baselines in terms of representation efficiency, and show interesting visual results for discretizing the maps.

Machado et al. describe a query system built on top of Cicero, a mixed-initiative game design assistant, that allows users to search for the occurence of specific events in human and generated gameplay replays. They present the results of a user study, showing that users were able to complete the assigned tasks of finding issues in a test game design, and that the tool was well-received by users.

Strategic Features for General Games by Cameron Browne, Dennis Soemers and Eric Piette

Browne et al. describe their approach to identify strategic, general features to bias Monte Carlo Tree Search rollouts for automated self-play of digital board games. The paper describes the descriptions of these features, implementation, and benefits of these features (such as Explainable AI).

Zhan et al. introduce the problem of automatically exploring video games, with the goal of extracting useful information at the same scale as human playtesting. They introduce many attempts at this problem, comparing them to human playtesters, and finding promising results (e.g. such as the completion of the first level in Super Mario World without costly reinforcement learning).

Rafati and Noelle present a framework for improving hierarchical reinforcement learning game playing agents via estimating subgoals (landmarks on the path to the goal) and rewarding attaining these subgoals.

Sturtevant explores a particular design aesthetic presented by Jonathan Blow and Marc ten Bosch. This design aesthetic requires designers to fully explore a possibility space, for example all the possible puzzles given a set of puzzle game mechanics. Therefore, Sturtevent employs a type of procedural content generation, Exhaustive Procedural Content Generation to fully explore and summarize possibility spaces for puzzles like those from The Witness.

Summary: Lin et al. describe an approach to generate a playable rhythm action game stage (called a chart) from an arbitrary piece of music, by deriving design knowledge through a deep neural network trained on existing charts.