Code for the preprint Learning State Abstractions for Transfer in Continuous Control by Kavosh Asadi, David Abel, and Michael L. Littman.
To run experiments:
> python run_learning_experiments_(domain).py
Where "(domain)" can be each of {puddle, lunar, cartpole}. To run the transfer experiments, open the file and set params['multitask'] to True.
To reproduce figure 3b, uncomment the call to num_training_data_experiment() in run_learning_experiments_puddle.py and run it.