subgoal-based-reward-shaping

This is for implementations of Subgoal-based Reward Shaping.

Setup

This implementation requires external packages, and you first install the packages by pip or other package management tools.

pip install -r requirements

Run

You can run a learning agent in each domain.

Navigation in a four-room domain

python -m fourroom --config=fourroom/in/configs/fourooms.ini

Navigation in a pinball domain

python -m pinball --config=pinball/in/configs/pinball.ini

Fetch robot picks and places

mpiexec -n 8 python -m picknplace --config=picknplace/in/configs/picknplace.ini

Config file

We place a config file per algorithm and label the file the name of algorithm like sarsa. If you want to try LEARNED with HUMAN, you selectlearned_human.

Analysis

The program for analysis is here.