<![CDATA[Artificial intelligence (AI) plays an important role in the gaming industry, its application commonly found in non-player character (NPC) and procedural content generation (PCG). One of the popular methods for developing AI is reinforcement learning (RL). Unity, as one of the most dominant game engines, has its own RL framework called Unity ML-Agents Toolkit. Unity with its capabilities to simulate realistic environments. allowing Unity ML-Agents Toolkit to develop and test RL agents in various complex scenarios. However, Unity ML-Agents Toolkit only has limited RL algorithms. This study aims to introduce an alternative method for implementing reinforcement learning in Unity, while addressing the Unity ML-Agents Toolkit’s limitations. The proposed method is Unity-Gymnasium, which integrates Unity with Gymnasium, and tested by developing a sliding puzzle environment. The result of this study demonstrates that the Unity-Gymnasium method works well and allows access to total 38 different RL algorithms from various RL libraries that compatible with Gymnasium, such as Stable Baseline 3, CleanRL, Tianshou, Ray Rllib, and Dopamine, this number is significantly higher compared to Unity ML-Agents Toolkit which only offer five RL algorithm options.]]>
Copyrights © 2025
