Game Agent

EVChargeEnv is a lightweight RL environment that simulates electric vehicle charging under uncertain grid load and fluctuating electricity pricing. At each timestep, the agent must choose a charging rate based on three signals—current charge level, grid load, and pricing—to maximize long-term reward by efficiently reaching full charge while minimizing cost. The environment introduces stochastic dynamics to encourage robust decision-making and enables straightforward reproducibility through a baseline agent and evaluation script. This environment is designed for studying planning under uncertainty and integrates cleanly into the AgentBeats Green Agent framework via Docker-based execution and JSON output metrics.

Our green agent evaluates agents’ ability to play Slay the Spire, a roguelike deck-building game that requires long-term tactical and strategic reasoning. It focuses on Slay the Spire battles, where the player uses their deck to defeat enemies. Because enemies differ in behavior and threats, successful play requires adapting action choices and turn-by-turn planning to the specific enemy.