---
tags:
- deep-reinforcement-learning
- reinforcement-learning
- decision-transformer
- gym-continous-control

pipeline_tag: reinforcement-learning

---
# Decision Transformer model trained on medium-replay trajectories sampled from the Gym HalfCheetah environment
This is a trained [Decision Transformer](https://arxiv.org/abs/2106.01345) model trained on medium-replay trajectories sampled from the Gym HalfCheetah environment

<video src="https://huggingface.co/edbeeching/decision-transformer-gym-halfcheetah-medium-replay/resolve/main/halfcheetah-medium-replay.mp4" controls autoplay loop></video>

The following normlization coeficients are required to use this model:

mean = [-0.12880704,  0.37381196, -0.14995988, -0.23479079, -0.28412786, -0.13096535, -0.20157982, -0.06517727,  3.4768248,  -0.02785066, -0.01503525,  0.07697279,  0.01266712,  0.0273253,   0.02316425,  0.01043872, -0.01583941] 

std = [0.17019016, 1.2844249,  0.33442774, 0.36727592, 0.26092398, 0.4784107, 0.31814206 ,0.33552638, 2.0931616,  0.80374336, 1.9044334,  6.57321, 7.5728636,  5.0697494,  9.105554,   6.0856543,  7.253004,  5]

See our [Blog Post](https://colab.research.google.com/drive/1K3UuajwoPY1MzRKNkONNRS3gS5DxZ-qF?usp=sharing),  [Colab notebook](https://colab.research.google.com/drive/1K3UuajwoPY1MzRKNkONNRS3gS5DxZ-qF?usp=sharing) or [Example Script](https://github.com/huggingface/transformers/tree/main/examples/research_projects/decision_transformer) for usage.