Game Agent using Genetic Algorithm

Abstract

The Mario AI Project is a pioneering project that uses genetic algorithms to teach an AI agent to play Super Mario. Inspired by natural selection, the algorithms work iteratively over generations to improve the AI agent’s gameplay strategies. Every generation sees the agent’s decision-making processes like jumping and navigating obstacles guided by a unique genetic code carefully crafted for optimum performance. Simulating natural Discrimination, Interchange and Transmutation, the strategy of AI Evolution agent undergoes a constant improvement.

Introduction

I. INTRODUCTION

In Mario AI, genetic Algorithmize AI entities that evolve over multiple generations. First, a population of potential fixes is created at random. Every solution depicts an AI agent with distinct characteristics. Fitter solutions are chosen for replication through evaluation according to fitness parameters like distance travelled and challenges conquered. Offspring that undergo crossover and mutation are more diverse and varied. Over several generations, this cyclical process of mutation, reproduction, and selection results in the creation of AI beings that are ever more skilled. In the end, genetic algorithms make it possible to create intelligent agents who can play the challenging platformer game Super Mario Bros. and navigate and complete levels on their own.

Artificial Intelligence (AI) has advanced significantly in the last several years, especially in the gaming industry. The creation of intelligent beings that can play video games on their own is an intriguing use of AI. Because of its rich intricacy and clear rules, the beloved platformer game "Super Mario Bros." Embarked a prominent testbed for AI research among these titles.

Genetic algorithms  are a method for developing AI bots that can play Super Mario Bros. The process of natural selection  refinement and evolution, in which solutions to problems develop over many generations, serves as an inspiration for genetic algorithms.

II. RELATED WORK

A. Rule-Based Systems

In Mario AI, rule-based systems use pre-established heuristics and rules to control how AI agents behave in the game world. Developers usually create these rules using their knowledge of the game mechanics and winning gameplay techniques. Rules could control things like dodging obstacles, hopping over opponents, and gathering power-ups. Rule-based systems have the advantage of being transparent; because the rules are clearly established, it is easy to comprehend how the AI agent behaves.

Moreover, such systems can be computationally efficient, as they do not require extensive training or learning processes. However, the Advancement of comprehensive rule sets for complex games like Mario Bros. can be challenging and time-consuming. Additionally, rule-based systems may struggle to adapt to novel or the Unanticipated scope of their predefined rules, limiting their flexibility and robustness in dynamic game environments.

B. Reinforcement Learning

Real-life agents engage with the gaming environment by acting and getting feedback in the form of incentives or punishments according to how well they succeed. Through frequent encounters, RL agents eventually pick up efficient techniques for completing objectives like clearing a level, gathering coins, and vanquishing foes. The Pivotal of reinforcement learning is its adaptability; RL agents can effectively navigate complicated and dynamic game contexts because they can adjust to new circumstances and learn from past mistakes. To converge to optimal behavior, RL agent training can be computationally demanding and necessitate Abundance interactions with the environment. Additionally, RL agents have to balance the exploration-exploitation trade-off, which can be difficult in unpredictable or stochastic environments like Super Mario Bros., between exploring novel actions and utilizing tried-and-true high-reward tactics.

III. METHODOLOGY

The first step in using the genetic algorithm with the PyBoy emulator for Mario AI is to initialize a population of possible solutions, each of which represents an AI agent Endowed with particular parameters that determine its behaviour. Via Py Boy, these agents communicate with the gaming environment and offer a simulated gaming platform. After the emulator has been initialized, Aptly they performed by looking at data like distance traveled, coins collected, and obstacles overcome. Then, selection techniques, which imitate natural selection principles, decide which agents advance to the next generation based on their fitness scores. By transferring genetic information across chosen agents to produce children with a combination of features, crossover promotes genetic diversity. Furthermore, mutation procedures add more variation to the population by introducing random modifications to its genetic makeup. As new offspring are introduced, the replacement mechanism makes sure that the population size stays constant. The performance of AI agents gradually increases through repeating cycles of crossover, mutation, and selection, convergent towards ideal solutions for traversing the complex and dynamic obstacles of Super Mario Bros. Through the Fusion PyBoy, researchers can effectively investigate an extensive array of AI agent setups, customizing solutions to thrive in the intricate and constantly changing Super Mario Bros. gaming environment.

Conclusion

To sum up, for this project, we created An independent Mario using a genetic algorithm. In conclusion, the Mario system has advanced significantly Consequence genetic algorithm\'s integration. The AI characters have performed better throughout the course of multiple generations, demonstrating the algorithm\'s ability to develop techniques that work. Notably, the AI can handle shifting settings and level designs with ease thanks to the algorithm\'s versatility. Trade-offs were made Concerning to processing resources, yet the AI is resilient Amidst variety of obstacles. The GE is positioned as a flexible and reliable method by comparative analysis. In order to further enhance the AI\'s behaviour in gaming scenarios, there are potential for future study to refine the algorithm, add new features, and investigate alternate techniques. This work represents a significant advancement in Proliferation intelligent and flexible AI systems for use in gaming. For instance, we did not create the condition in our genetic process where Mario might pick flowers and mushrooms. Furthermore, our system prioritizes optimizing the success rate. Proposition useful overview of this issue and will be helpful to anyone who are interested in playing computer games with genetic algorithms.

References

[1] Mario AI Championship 2010: Results (2010).https://sites.google.com/a/marioai.com/www/results. Accessed 24 May 2015 [2] Bellemare, M.G., Naddaf, Y., Veness, J., Bowling, M.: The arcade learning environment: an evaluation platform for general agents. JAIR 47, 253–279 (2013) [3] Emgallar: Intelligent NPC for Mario AI Championship(2011).https://www.youtube.com/watch?v=u 0pgFQ8HcM. Accessed 22 May 2015 [4] Fujii, N., Sato, Y., Wakama, H., Katayose, H.: Autonomously acquiring a video game agent’s behavior: letting players feel like playing with a human player. In:Nijholt, A., Rom˜ao, T., Reidsma, D. (eds.) ACE 2012. LNCS, vol. 7624, pp. 490–493. Springer, Heidelberg (2012) [5] Karakovskiy, S., Togelius, J.: The Mario AI benchmark and competitions. IEEETrans. Comput. Intell. AI Games 4(1), 55–67 (2012) [6] Meffert, K., Rotstan, N.: JGAP: Java Genetic Algorithms Package (2015). http://jgap.sourceforge.com. Accessed 6 July 2015 [7] Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A., Veness, J., Graves, A., Riedmiller,M., Fidjeland, A., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A., Antonoglou, I.,King, H., Kumaran, D., Wierstra, D., Legg, S., Hassabis, D.: Human-level controlthrough deep reinforcement learning. Nature 518, 529–533 (2015) [8] Steadman, I.: This AI ‘Solves’ Super Mario Bros.andOtherClassicNESGames(2013).http://www.wired.co.uk/news/archive/2013-04/12/super-mariosolved. Accessed 22 May 2015 [9] Togelius,J.,Karakovskiy, S., Baumgarten, R.: The 2009 Mario AI competition. In:2010 IEEE Congress on Evolutionary Computation, pp. 1–8 (2010) [10] Markus Persson. Infinite Mario Bros. http://www.mojang.com/notch/mario/. 2011. [11] Jarret Raim. Finite State Machine In Game,from:www.cse.lehigh.edu/~munoz/CSE497/classes/FSM_In_Games.ppt. 2011. [12] K.-T. Chang, K.-O. Chin, and J. Teo, “The Evolution of Gamebots for 3D First Person Shooter (FPS),” in press. The Sixth International Conference on Bio-Inspired Computing: Theories and Applications, 2011. [13] A.E. Eiben, and J.E. Smith, Introduction to Evolutionary Computing. Springer, 2003

Copyright

Copyright © 2024 V Akshit Rao, P Sai Kiran, Prateet M, K Sudhakar Reddy. This 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.