ToppleTim | Devpost

Happy Tim
Running Tim
Topple Tim

Inspiration

We wanted to create a game that was not only fun but also smart about handling challenges. Our goal was to design a game where every move matters, and risk is carefully managed. We were also excited about using reinforcement learning to help make the game more dynamic and balanced.

What it does

ToppleTim is a game where your only goal is to stop Tim from reaching his destination by toppling him over. In each level, you throw balls at Tim, trying to knock him down. It’s all about planning your moves and managing risk, as every action can lead to unexpected results. The game combines realistic physics with smart AI using reinforcement learning along with a touch of randomness to adjust to throws on the fly, ensuring that every level is both engaging and fair.

How we built it

Unity, Unity-MLAgents, Blender

Challenges we ran into

Configuring reinforcement learning was the most challenging part—it involved modeling the entire scenario, figuring out how to train the model, defining the state space, and setting up the appropriate constraints and rewards.

Accomplishments that we're proud of

None of us knew Blender or Unity, I think building an entire game from scratch is something, we as a team, are extremely proud of.

What we learned

Through this project, we discovered the vital role of rapid iteration and teamwork. We also realized that game development is complex—with many moving parts and challenging UI design—but it’s equally rewarding.