My Tamagotchi Is an RL Agent Playing Slither.io (nkasmanoff.github.io)

A developer has created a reinforcement learning (RL) agent that plays the multiplayer game Slither.io using a Raspberry Pi. Inspired by OpenAI's Gym and the nostalgia of training AI agents, the project aims to develop a bot that mimics human behavior in the game by efficiently navigating and avoiding obstacles while maximizing food consumption. Utilizing tools like Selenium and JavaScript injection, the agent effectively extracts game state data to inform its actions. The project underwent several iterations, exploring different methods including REINFORCE and the more sophisticated Advantage Actor-Critic (A2C) algorithms to improve decision-making. This project is significant for the AI/ML community as it showcases practical applications of RL techniques in real-time environments. It highlights challenges such as balancing training efficiency with gameplay speed by implementing a network architecture and action space constrained for rapid response. The successful agent was able to learn to navigate while avoiding danger and optimizing food collection, thereby demonstrating how RL can be effectively applied in dynamic scenarios. The experiments underscore the complexities involved in shaping reward functions and highlight the importance of balancing immediate rewards with long-term survival goals, contributing valuable insights into the training of RL agents.