My participation in OrbisChallenge 2014 (which ran for a length of a 24-hour period), in which I had to supply a driver in Tron to compete with the AI, and potentially to compete against AIs of different strategies.
My approach was to reduce the game to the (commonly known in AI as the) "search problem", and applying the minimax rule for minimizing potential loss of outcome of a series of decisions for consecutive moves (within the evaluation threshold, where both players get to move in one direction for each decision frame) for an instance of the game.
A majority of representing a configuration of Tron had already been implemented in the starter code, but due to the simplicity of the game there is little to no deviation in representations of a configuration of state of a Tron game for optimizing decision procedures.
- Document my approach more throughly, with references to the code
- Fix some bugs in my implementation of minimax
- Determine a favourable heuristic,
h(n)for IDA* graph traversal - Add alpha-beta pruning to execution of minimax for performance improvement
- Modularize decision rules for the decision procedure so that when the user goes to run the simulation from the command-line, they can choose which set of rules their driver takes for that instance of the game
- Add a smart minimax bound whereby a maximum depth is determined at the beginning such that each branch of the search problem is bounded by that depth to satisfy a global time constraint.
- Add efficient bulk loading of multiple simulations