Views
No views yet
pretrained_checkpoints/
├── DMPO_pretrained_gym_checkpoints/
│ ├── gym_improved_meanflow/ # MeanFlow without dispersive loss
│ └── gym_improved_meanflow_dispersive/ # MeanFlow with dispersive loss (recommended)
└── DMPO_pretraining_robomimic_checkpoints/
├── w_0p1/ # dispersive weight = 0.1
├── w_0p5/ # dispersive weight = 0.5 (recommended)
└── w_0p9/ # dispersive weight = 0.9| Domain | Tasks |
|---|---|
| OpenAI Gym | hopper, walker2d, ant, humanoid, kitchen-* |
| Robomimic (RGB) | lift, can, square, transport |
hf:// prefix in config files to auto-download:1# Gym tasks
2base_policy_path: hf://pretrained_checkpoints/DMPO_pretrained_gym_checkpoints/gym_improved_meanflow_dispersive/hopper-medium-v2_best.pt
3
4# Robomimic tasks
5base_policy_path: hf://pretrained_checkpoints/DMPO_pretraining_robomimic_checkpoints/w_0p5/can/can_w0p5_08_meanflow_dispersive.pt1@misc{zou2026stepenoughdispersivemeanflow,
2 title={One Step Is Enough: Dispersive MeanFlow Policy Optimization},
3 author={Guowei Zou and Haitao Wang and Hejun Wu and Yukun Qian and Yuhang Wang and Weibing Li},
4 year={2026},
5 eprint={2601.20701},
6 archivePrefix={arXiv},
7 primaryClass={cs.RO},
8 url={https://arxiv.org/abs/2601.20701},
9}