/Meta-MDP

Primary LanguagePython

A Meta-MDP Approach to Exploration for Lifelong Reinforcement Learning

  • To train the exploration policy use LearnMeta.py. For example for carpole:

    • python LearnMeta.py -e cartpole -d ./CartpoleData/cartpole_data.pkl -a 0.001 -b 0.010 -s ./MetaTrainingCartpole (Note: for cartpole and animat we included pre-recorded data to simplify learning. It is not necessary to train the advisor)

  • To get a baseline of a random exploration use RandomBaseline.py. Example:

    • python RandomBasline.py -e cartpole -d ./CartpoleData/cartpole_data.pkl -s ./RandomTrainingCartpole

Unfortunately, due to upload size we are not able to include the necessary files for the self-driving domain. The changes to roboschool for creating model variations, require modifying the .xml model files after installation.