devdharpatel/SAC-Humanoid-v2

Soft-Actor-Critic: Humanoid-v2

These are 25 trained models over seeds (0-4) and J = 1, 2, 4, 8, 16 of Soft actor critic agent playing Humanoid-v2 for Sequence Reinforcement Learning (SRL).

Model Sources

Repository: https://github.com/dee0512/Sequence-Reinforcement-Learning
Paper (ICLR): https://openreview.net/forum?id=w3iM4WLuvy
Arxiv: arxiv.org/pdf/2410.08979

Training Details:

Using the repository:

python .\train_sac.py --env_name <env_name> --seed <seed> --j <j>

Evaluation:

Download the models folder and place it in the same directory as the cloned repository. Using the repository:

python .\eval_sac.py --env_name <env_name> --seed <seed> --j <j>

Metrics:

FAS: Frequency Averaged Score
j: Action repetition parameter

Citation

The paper can be cited with the following bibtex entry:

BibTeX:

@inproceedings{DBLP:conf/iclr/PatelS25,
  author       = {Devdhar Patel and
                  Hava T. Siegelmann},
  title        = {Overcoming Slow Decision Frequencies in Continuous Control: Model-Based
                  Sequence Reinforcement Learning for Model-Free Control},
  booktitle    = {The Thirteenth International Conference on Learning Representations,
                  {ICLR} 2025, Singapore, April 24-28, 2025},
  publisher    = {OpenReview.net},
  year         = {2025},
  url          = {https://openreview.net/forum?id=w3iM4WLuvy}
}

APA:

Patel, D., & Siegelmann, H. T. Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control. In The Thirteenth International Conference on Learning Representations.