Implementation of Deep Deterministic Policy Gradient Algorithm in Tensorflow to play Torcs
To train or Test Modify: my_config.py run: python playGame_tensorflow.py
Usually you should observe high rewards for around ~300th or 400th episode. If you are still getting low rewards. Please restart training.
For faster training avoid the complete rendering of the graphics. This can be done by replacing the practice.xml to practice_results_only.xml
There is a configuration file "practice.xml" that exists in one of the 2 folders which can be found by running the command
sudo find / -name practice.xml
One of the folders is "~/.torcs/config/raceman" Other can be found as specified.
Torcs can run in 2 modes either the complete race simulation or results only <10X faster> For training I replace practice.xml by practice_results_mode.xml For visualization again replace practice.xml by practice_normal_mode.xml