amiroor / rl_helicoopter

Our old implementation of Reinforcement Learning for a helicoopter game. Learning is by soft RBF regression!

Reinforcement Learning in Helicopter Game

This is a proof of concept RL implementation for a helicopter game. The helicopter learns to avoid obstacles by learning a reward function based on Radial Basis Clustering.

Compile

You should have qt4 or qt5 and a c++ compiler. Type the following commands in shell:

qmake
make

Usage

Run the runnable output of make. Then,

    Press
        <space> to pause/resume
        <S>     to save clusters
        <P>     to switch to autpilot mode
        <I>     to save episode
        <L>     to load clusters
        <G>     to start learning in background mode
        <R>     to switch to replay mode (then use forward and backward to jump 1000 episodes forward or backward

Train on background and replay

1. Press 'I' to enable saving episodes. They will be saved in './episodes/' folder.
2. Press 'G' to run background training and wait until finishing...
3. When it is done, press 'R' to replay. press left/right arrow keys to move 1000 episodes back/forward in the player.