Unsupervised learning of Goal Spaces for Intrinsically Motivated Exploration

This repository hosts the code to reproduce the results presented in the paper Unsupervised Learning of Goal Spaces for Intrinsically Motivated Goal Exploration. In this paper, we propose a novel exploration algorithmic architecture that uses a goal space learned using representation learning algorithms. Experiments are performed on two simple tasks in which multi-joint arm must handle and gather an oobject in a 2D space.

Running the experiments

To run a single experiment, you can run one of the three following python scripts:

• rpe.py to perform a Random Parameterization Exploration
• rge_efr.py to perform a Random Goal Exploration using Engineered Features Representation
• rge_rep.py to perform a Random Goal Exploration using a learned Representation

You can also run a full campaign batch by executing campaign.sh.

Finally, to generate the different figures out of the raw results, you can use the two notebooks:

• Experiment_Visualization.ipynb to visualize the data of a single run
• Performance_Comparison.ipynb to visualize the compared performance of multiple runs