Inspiration
The difficulty and the lack of a conclusive solution inspired us, as we were forced to experiment with multiple approaches and see what stuck.
What it does
The project effectively works as an intelligent computer player of a game very similar to Tron. It uses several techniques to forecast and score the available moves based on their features(e.g. how much open space does following the square provide the bot) and ability to produce victories deeper in the game.
How we built it
The agent was built with a recursive Minimax algorithm, applying alpha-beta pruning and custom heuristics to forecast each move’s risks and benefits under tight board and time constraints. All board state logic, move simulation, and evaluation features are organized into efficient Python modules for reliable and rapid decision-making.
Challenges we ran into
Tuning the search algorithm and evaluation heuristics to avoid dead ends and ensure the agent reliably outmaneuvered adversaries was complex and required extensive debugging.
Accomplishments that we're proud of
We achieved a bot capable of deep, adaptive forecasting in real time, able to survive and win against competitive opponents.
What we learned
Incremental improvements in heuristics and stability are crucial, and success comes from iterative testing and practical adjustment.
What's next for Agents
We plan to explore learning-based heuristics, stronger opponent modeling, and more advanced search techniques for even deeper foresight.
Log in or sign up for Devpost to join the conversation.