The Many Challenges of Human-Like Agents in Virtual Game Environments

@inproceedings{humanLike2025,
  title={{The Many Challenges of Human-Like Agents in Virtual Game Environments}},
  author={{\'S}wiechowski, Maciej and {\'S}l{\k{e}}zak, Dominik},
  booktitle={Proc. of the 24th International Conference on Autonomous Agents and Multiagent Systems (AAMAS'25)},
  year={2025},
  doi={10.5555/3709347.3743837},
  isbn = {9798400714269},
  publisher={IFAAMAS},
  pages={1996--2005},
  series = {AAMAS '25}
}
  

This work explores the core difficulties of developing and evaluating believable -- human-like agents in virtual game environments.

It consists of two parts. The first part is a survey identifying 13 key challenges through analysis of 54 research papers.

The second part presents an experiment in a squad-based tactical game, where a machine learning model is used to differentiate between human and bot players. The approach utilizes a Deep Recurrent Convolutional Neural Network.

Let's assume that for certain types of bots - the model is very accurate. Our idea is to then use this model to create more human-like agents by evaluating them based on how well they can fool the detector. This is a novel creation pipeline to creating believable bots.