Rinascimento: using event-value functions for playing Splendor

TL;DR

This paper introduces Rinascimento, an approach that uses event-value functions based on game state changes to improve AI performance in Splendor, especially when point rewards are sparse or delayed.

Contribution

It proposes a novel event logging and value assignment method that enhances AI learning and strategic control in games with scarce reward signals.

Findings

Event-value functions mitigate sparse reward issues

Enhanced AI robustness against multiple opponents

Richer behavioral control through event-based strategies

Abstract

In the realm of games research, Artificial General Intelligence algorithms often use score as main reward signal for learning or playing actions. However this has shown its severe limitations when the point rewards are very rare or absent until the end of the game. This paper proposes a new approach based on event logging: the game state triggers an event every time one of its features changes. These events are processed by an Event-value Function (EF) that assigns a value to a single action or a sequence. The experiments have shown that such approach can mitigate the problem of scarce point rewards and improve the AI performance. Furthermore this represents a step forward in controlling the strategy adopted by the artificial agent, by describing a much richer and controllable behavioural space through the EF. Tuned EF are able to neatly synthesise the relevance of the events in the game. Agents using an EF show more robust when playing games with several opponents.