Views
No views yet
1README.md
2libero_uncond_2cam224.pt
3libero_uncond_2cam224_dataset_stats.json
4libero_optional_idm_2cam224.pt
5libero_optional_idm_2cam224_dataset_stats.json
6robotwin_uncond_3cam_384.pt
7robotwin_uncond_3cam_384_dataset_stats.jsonlibero_uncond_2cam224.pt: FastWAM Base for
task=libero_uncond_2cam224_1e-4.libero_optional_idm_2cam224.pt: Optional IDM for
task=libero_optional_idm_2cam224_1e-4.robotwin_uncond_3cam_384.pt: FastWAM Base for
task=robotwin_uncond_3cam_384_1e-4.*_dataset_stats.json file for evaluation.idm: imagine the future video first, then predict actions.first_frame: predict actions directly from the current observation without
test-time future imagination.1python experiments/libero/run_libero_manager.py \
2 task=libero_optional_idm_2cam224_1e-4 \
3 ckpt=./checkpoints/fastwam_release/libero_optional_idm_2cam224.pt \
4 EVALUATION.dataset_stats_path=./checkpoints/fastwam_release/libero_optional_idm_2cam224_dataset_stats.json \
5 EVALUATION.sigma_shift=1.0 \
6 +EVALUATION.action_infer_mode=idm \
7 MULTIRUN.num_gpus=8idm with first_frame to use the Fast-WAM inference mode. The
released checkpoint uses an action scheduler shift of 1.0 for training and
evaluation.| Inference mode | Spatial | Goal | Object | Long | Average |
|---|---|---|---|---|---|
| IDM | 99.0% | 98.6% | 99.6% | 97.0% | 98.55% |
| First frame (Fast-WAM) | 98.2% | 97.8% | 99.2% | 95.8% | 97.75% |
1@misc{yuan2026fastwam,
2 title={Fast-WAM: Do World Action Models Need Test-time Future Imagination?},
3 author={Tianyuan Yuan and Zibin Dong and Yicheng Liu and Hang Zhao},
4 year={2026},
5 note={arXiv preprint arXiv:2603.16666}
6}