Interactive Learning of Temporal Features for Control
Code of the paper "Interactive Learning of Temporal Features for Control" published in the IEEE Robotics & Automation Magazine (Special Issue on Deep Learning and Machine Learning in Robotics).
This code is based on the following publication:
- Interactive Learning of Temporal Features for Control, preprint availabe here.
Authors: Rodrigo Pérez-Dattari, Carlos Celemin, Giovanni Franzese, Javier Ruiz-del-Solar, Jens Kober.
This repository includes the code necessary to run the experiments done in simulated environments using human teachers.
To use the code, it is necessary to first install the gym toolkit (release v0.9.6): https://github.com/openai/gym
Then, the files in the
gym folder of this repository should be replaced/added in the installed gym folder in your PC. Two environments were added:
Continuous Mountain Car: the environment outputs an image as an observation.
Inverted Pendulum: the pendulum is bigger and the environments outputs an image as an observation.
- To run the main program type in the terminal (inside the folder
python main.py --config-file <config_file_name>
To be able to give feedback to the agent, the environment rendering window must be selected/clicked.
This code has been tested in
Ubuntu 18.04 and
python >= 3.5.
If you run into problems of any kind, don't hesitate to open an issue on this repository. It is quite possible that you have run into some bug we are not aware of.