Inspiration
We wanted to push the boundaries of AI in gaming by integrating machine learning into a Mario-style platformer. The idea was to train an AI to learn from player behavior and improve over time, making the game more dynamic and engaging.
What it does
Our game is a fully functional platformer with classic mechanics—running, jumping, and navigating obstacles. While our initial goal was to integrate AI using ML-Agents or NEAT, we faced challenges that prevented full implementation within the hackathon timeframe.
How we built it
We built the game entirely from scratch in Unity, designing levels, implementing player movement, and refining mechanics for smooth gameplay. To enhance visuals, we used ShaderLab and HLSL to create custom shaders, improving lighting, textures, and overall aesthetics. Alongside this, we explored AI training methods and real-time data collection to prepare for future integration.
Challenges we ran into
The biggest hurdle was integrating AI into our existing Unity game. We faced compatibility issues, ML-Agents installation problems, and difficulties in setting up real-time communication between Unity and Python. Additionally, working with ShaderLab and HLSL required trial and error to achieve the desired visual effects. Due to time constraints, we couldn't complete the AI training as planned.
Accomplishments that we're proud of
Despite the setbacks, we built a polished game with a solid foundation for AI-driven enhancements. We successfully implemented custom shaders using HLSL, created a visually appealing game world, and gained hands-on experience with Unity’s rendering pipeline.
What we learned
We learned a lot about building a game from scratch in Unity, including working with ShaderLab and HLSL to create visual effects. We also explored AI integration with ML-Agents, which gave us insight into reinforcement learning and how AI can interact with game environments. While we faced challenges in fully integrating the AI, we gained a deeper understanding of Unity’s systems, debugging complex issues, and managing project deadlines. This experience has shown us the importance of planning ahead for AI implementation and has motivated us to keep improving our skills.
What's next for How to Train Your AI
After the hackathon, we plan to revisit AI integration, properly setting up ML-Agents or NEAT to train an intelligent agent. We’re also interested in refining our shaders further to enhance visual fidelity.
Log in or sign up for Devpost to join the conversation.