Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{206360,
        author = {Sachin Damre},
        title = {Implementation and Performance Evaluation of Curriculum Learning and Behavior Cloning for Multi-Agent Freeze Tag Environment using Unity ML-Agents},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {2},
        pages = {1394-1401},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=206360},
        abstract = {Recent advances in Reinforcement Learning (RL) have enabled autonomous agents to learn complex decision-making strategies through continuous interaction with dynamic environments. Among the various training methodologies, Curriculum Learning (CL) and Behavior Cloning (BC) have gained significant attention for improving learning efficiency and policy quality in multi-agent systems. Curriculum Learning adopts a progressive training strategy by gradually increasing task complexity, allowing agents to acquire foundational skills before solving more challenging scenarios. In contrast, Behavior Cloning leverages expert demonstrations to directly imitate desired behaviours, thereby reducing the exploration burden during the initial stages of training.
This paper presents the design, implementation, and comparative evaluation of Curriculum Learning and Behavior Cloning within a custom-developed 3v3 Freeze Tag environment implemented using the Unity ML-Agents framework. The environment models a cooperative-competitive game in which runner agents attempt to survive and rescue teammates while tagger agents seek to freeze all opposing agents within a predefined time limit. The proposed implementation integrates Proximal Policy Optimization (PPO) for reinforcement learning and combines Behavior Cloning with Generative Adversarial Imitation Learning (GAIL) to enhance policy learning from expert demonstrations.},
        keywords = {Reinforcement Learning, Deep Reinforcement Learning, Unity ML-Agents, Curriculum Learning, Behavior Cloning, Generative Adversarial Imitation Learning, Proximal Policy Optimization, Freeze Tag Environment, Multi-Agent Systems, Artificial Intelligence.},
        month = {July},
        }