Normalizing Flows are Capable Models for RL

📄 Paper

• Normalizing Flows are Capable Models for RL
  Raj Ghugare, Benjamin Eysenbach

🌐 Project Page

• Project Website
  Includes abstract and results.

Installation and usage

Separate instructions for each RL task can be found in the respective subfolders.

• Imitation learning and conditional imitation learning
• Offline RL
• Unsupervised goal conditional RL

Acknowledgments

We thank the respective authors for their awesome contributions.

• real-nvp-pytorch for a simple RealNVP implementation.
• ogbench for offline GCRL benchmarks and baseline code.
• fql for offline RL benchmarks and baseline code.
• tarflow for a neat normalizing flow codebase.