Zijie-Xu/ANN-SNN_Continuous_Control

Error Amplification Limits ANN-to-SNN Conversion in Continuous Control

Official model release for the ICML 2026 paper [arXiv] [GitHub]

Well-trained ANN Models

The models for DDPG/TD3/SAC agents trained on MuJoCo for 3M steps are stored in ./DDPG&TD3&SAC_MuJoCo/models.

The models for DrQ-v2 agents trained on DMC are stored in ./DrQ-v2_DMC/exp_local.

More Details

Code is available at this GitHub repository. You can also view our paper PDF.

Citing This

@article{xu2026error,
  title={Error Amplification Limits ANN-to-SNN Conversion in Continuous Control},
  author={Xu, Zijie and Huang, Zihan and Dong, Yiting and Chen, Kang and Liu, Wenxuan and Yu, Zhaofei},
  journal={arXiv preprint arXiv:2601.21778},
  year={2026}
}