Built a robocode tank (a Reinforcement Learning agent) that battles with another robot (an enemy) and defeats it ~80% of the time. The error backpropagation algorithm as applied to multi-layer perceptron and reinforcement learning is used for the robot’s attack or defence strategy.
Before training the battle looks something like this:
- Yellow battle tank : Opponent
- Blue battle tank : My battle tank (Not trained)
After training:
- Yellow battle tank : Opponent
- Blue battle tank : My battle tank (Trained)