All You Need Is Supervised Learning: From Imitation Learning to Meta-RL With Upside Down RL
Implementation of a generalised UDRL agent. Supports online RL, imitation learning, offline RL, goal-conditioned RL, and meta-RL:
python main.py --mode [online|imitation|offline|goal|meta]Note that some implementation decisions were made for simplicity, and hence do not reflect an implementation that works across all modes simultaneously.
Requirements can be installed with:
pip install -r requirements.txt| Online | IL | Offline | GCRL | Meta-RL |
|---|---|---|---|---|
 |
 |
 |
 |
 |
@inproceedings{arulkumaran2022all,
author = {Arulkumaran, Kai and Ashley, Dylan R. and Schmidhuber, Jürgen and Srivastava, Rupesh K.},
title = {All You Need is Supervised Learning: From Imitation Learning to Meta-RL with Upside Down RL},
booktitle = {Multi-disciplinary Conference on Reinforcement Learning and Decision Making},
year = {2022}
}