dmux/CIR

Constrained Initial Representations

This is the pre-trained policy model weights for our novel method, Constrained Initial Representations (CIR).

The results are grouped into each environment and seed. One can directly load the policy weights from the corresponding directory

One can find our papers here

Citation

If you find our work interesting or use our work in your paper, please consider citing our paper:

@article{lyu2026temporal,
  title={Temporal Difference Learning with Constrained Initial Representations},
  author={Lyu, Jiafei and Yang, Jingwen and Qiao, Zhongjian and Liu, Runze and Liu, Zeyuan and Ye, Deheng and Lu, Zongqing and Li, Xiu},
  journal={arXiv preprint arXiv:2602.11800},
  year={2026}
}