This repository contains various systems for which the empowerment landscape can be computed.
python3 train/empowerment_train.py --env 0
A list of the sytems:
- 0: Pendulum
- 1: Ball in box
- 2: Robot arm
Empowerment can be computed with and without a filter that estimates hidden states.
python3 train/empowerment_train.py --use_filter 1
Finally, a policy maximizing empowerment can be trained.
python3 train/policy_train.py