Projects‎ > ‎

Learning Robocode Robot

A simulated robot that learns which strategy is most effective against a given enemy. Uses the reinforcement learning library of Dr Tim Kovacs (University of Bristol, department of Computer Science).

The robot has two methods. Using the first the robot learns which actions (turning, shooting, moving) lead to the best rewards. This did not work well due to the large number of actions. The second method was much more successful. For each round the robot randomly chose a strategy taken from an open source robot, and eventually learned which ones were most effective. This performed well against high ranking robots on a Robocode ladder.

Language: Java