joint_model

Overview

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