ПОРІВНЯННЯ НАВЧАННЯ З ПІДКРІПЛЕННЯМ В ІГРОВИХ РУШІЯХ  UNREAL ENGINE 5 ТА UNITY

Ivan Laitaruk; Tetiana Hryshanovych; Oksana Onyshchuk

Authors

Ivan Laitaruk Lesya Ukrainka Volyn National University https://orcid.org/0009-0002-9832-7759
Tetiana Hryshanovych Lesya Ukrainka Volyn National University https://orcid.org/0000-0002-3595-6964
Oksana Onyshchuk Lesya Ukrainka Volyn National University https://orcid.org/0000-0002-8342-3011

Keywords:

Unreal Engine 5 (UE5), Unity, Blueprints, C#, machine learning, agents, Reinforcement learning

Abstract

This paper presents a comparative analysis of reinforcement learning implementation in two leading game engines: Unreal Engine 5 and Unity. The study evaluates the integration simplicity, training effectiveness, and tool support of each platform by examining similar open-source projects where autonomous car agents learn to navigate racetracks using the Proximal Policy Optimization (PPO) algorithm. In Unreal Engine 5, training is achieved through the new Learning Agents plugin, leveraging Blueprint scripting and real-time physics-based simulation, while in Unity, the ML-Agents Toolkit is used with C# scripting and external Python-based training. The analysis includes environment setup, reward structures, agent perception systems, and debugging tools. Additionally, user experience and technical documentation quality are assessed. The results provide insights for developers and researchers aiming to choose an optimal platform for integrating RL into interactive simulations or games.

References

TD-Gammon / W. Uther et al. Encyclopedia of machine learning. Boston, MA, 2011. P. 955–956. URL: https://doi.org/10.1007/978-0-387-30164-8_813.

Gronauer S., Diepold K. Multi-agent deep reinforcement learning: a survey. Artificial intelligence review. 2021. URL: https://doi.org/10.1007/s10462-021-09996-w.

Egri-Nagy A., Törmänen A. The cost of passing - using deep learning AIs to expand our understanding of the ancient game of Go. International journal of networking and computing. 2023. Vol. 13, no. 2. P. 258–272. URL: https://doi.org/10.15803/ijnc.13.2_258.

Yadav A. Reinforcement Learning in Games: A Complete Guide. Medium. URL: https://medium.com/@amit25173/reinforcement-learning-in-games-a-complete-guide-24d1cab79317.

Risi S., Preuss M. Behind DeepMind’s AlphaStar AI that Reached Grandmaster Level in StarCraft II. KI - Künstliche Intelligenz. 2020. Vol. 34, no. 1. P. 85–86. URL: https://doi.org/10.1007/s13218-020-00642-1.

Welcome to AirSim. AirSim. URL: https://microsoft.github.io/AirSim/.

Mulcahy B. Learning agents (5.5). Unreal Engine Dev Community. URL: https://dev.epicgames.com/community/learning/courses/GAR/unreal-engine-learning-agents-5-5/7dmy/unreal-engine-learning-to-drive-5-5.

Mulcahy B. Learning Agents Introduction (5.3). Unreal Engine Dev Community. URL: https://dev.epicgames.com/community/learning/tutorials/8OWY/unreal-engine-learning-agents-introduction-5-3.

ML-Agents Overview. Unity ML-Agents Toolkit. URL: https://unity-technologies.github.io/ml-agents/ML-Agents-Overview/.

JulesVerny/MLAgentsAtSpa: Unity ML Agent Racing Car learns to Drive at Spa !. GitHub. URL: https://github.com/JulesVerny/MLAgentsAtSpa.