Kokomo is a competitor robot built for Robocode that uses regression tree based machine learning (called Dynamic Segmentation by some in the Robocode community) to learn its opponent's movement strategy.
Like most top competitor robots, Kokomo collects GuessFactors using Waves - both its own and its opponents. Movement uses traditional Wave Surfing (a slight improvement over BasicGFSurfer's reference Wave Surfing implementation). Targeting uses a custom incremental regression tree.
- Improve the performance of the regression tree and solve several memory issues that occur when matches go on too long
- Implement serialization, like described in the Wiki Targeting page, to improve early round performance against top bots
- Expand regression tree into a forest, starting with an implementation of a k-d tree and a k-nn targeting algorithm
- Investigate using machine learning for movement as well