Evolving unsupervised neural networks for Slither.io

Slither.io is a massively multiplayer online game in which up to 500 players control worm-like avatars and consume food to grow with the goal of becoming the largest player while avoiding running into one another. The platform serves as a good testbed for developing AI controlled agents due to its accessibility, mechanical simplicity, and unpredictability. In this paper, we develop a Slither.io bot using neuroevolution of augmenting topologies (NEAT) and compare its performance to that of the best open source bot available online (a high-performing expert system bot). With a fitness function based on the final size of the agent, our results show steady improvement in average score. We discuss the unique emergent behaviors observed by our top performing agents.

Paper

Full text

PDF

Evolving unsupervised neural networks for Slither.io

Semantic Scholar · Computer Science · 2019

Abstract

Slither.io is a massively multiplayer online game in which up to 500 players control worm-like avatars and consume food to grow with the goal of becoming the largest player while avoiding running into one another. The platform serves as a good testbed for developing AI controlled agents due to its accessibility, mechanical simplicity, and unpredictability. In this paper, we develop a Slither.io bot using neuroevolution of augmenting topologies (NEAT) and compare its performance to that of the best open source bot available online (a high-performing expert system bot). With a fitness function based on the final size of the agent, our results show steady improvement in average score. We discuss the unique emergent behaviors observed by our top performing agents.

Similar papers

© 2026 NYSGPT2525 LLC