An autonomous agent framework for the game "Don't Starve"
Resumo
Applying AI algorithms to find optimal strategies in games has been a heavily researched subject that had a significant impact on video game development and served as the basis for several applications in the real world. In this paper, we investigate developing an AI agent for a game called “Don’t Starve”, a single-player survival game. The main contribution of this work is a novel agent framework for this game that can survive under the same conditions as a human player. In this regard, the agent uses the game screen as the only source of information and executes actions with mouse and keyboard. After testing the AI capabilities, the agent could identify game objects correctly, plan its next steps, collect important resources, and survive for a few days.
Referências
Berner, C. et al. (2019). Dota 2 with large scale deep reinforcement learning. CoRR, abs/1912.06680.
Cartucho, J., Ventura, R., e Veloso, M. (2018). Robust object recognition through symbiotic deep learning in mobile robots. In 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 2336–2341.
da Silva, F. B. D. R., de Albuquerque Máximo, M. R. O., Yoneyama, T., Barroso, D. H. V., e Aki, R. T. (2024). Calibration of inverse perspective mapping for a humanoid robot. In Buche, C., Rossi, A., Simões, M., e Visser, U., editors, RoboCup 2023: Robot World Cup XXVI, pages 117–128, Cham. Springer Nature Switzerland.
Klei Entertainment (2024). Don’t starve. [link]. Acesso em: 10 de mai. de 2024.
Szeliski, R. (2010). Computer Vision: Algorithms and Applications. Springer-Verlag, Berlin, Heidelberg, 1st edition.
Ultralytics (2023). YOLOv8. [link]. Acesso em: 10 de mai. de 2024.
Vinyals, O. et al. (2019). Grandmaster level in starcraft ii using multi-agent reinforcement learning. Nature, pages 1–5.
