Published on: Feb 8,
2025
Welcome to the development blog for Sumo Trainer, an AI-based minigame where players train a character powered by a custom neural network. I’m Loui, a backend developer from Germany. This blog documents the project’s progression—from the initial idea to technical insights into AI implementation, game design decisions, and key learnings. New posts are published regularly, focusing on how imitation learning is applied to bring sumo wrestlers to life. A central aspect of Sumo Trainer is the use of imitation learning, which allows players to train their characters through demonstration. The focus is not only on match outcomes but also on the training process itself—observing behavior, evaluating performance, and iterating on improvements. The effectiveness of a trained wrestler depends on how well movement, reactions, and strategies have been demonstrated during training. This blog follows the ongoing development of the game and explores how machine learning techniques can be integrated into an interactive gameplay experience.