---
license: mit
pretty_name: Faster-WAM Checkpoints
library_name: pytorch
tags:
- robotics
- robot-learning
- imitation-learning
- embodied-ai
- libero
- robotwin
---

# Faster-WAM Checkpoints

This repository provides the official checkpoints for **[Faster-WAM: Efficient Inference-Time Future Conditioning for Robust World Action Models](https://arxiv.org/pdf/2608.04404)**.

## Files

```text
libero/
├── step_021700.pt
└── dataset_stats.json
robotwin/
├── step_029355.pt
└── dataset_stats.json
```

- `libero/step_021700.pt`: checkpoint for LIBERO and LIBERO-Plus evaluation.
- `robotwin/step_029355.pt`: checkpoint for RoboTwin evaluation.
- `dataset_stats.json`: dataset statistics required by the corresponding evaluation pipeline.

## Links

- Code: [https://github.com/hustvl/FasterWAM](https://github.com/hustvl/FasterWAM)
- Paper: [https://arxiv.org/pdf/2608.04404](https://arxiv.org/pdf/2608.04404)

## Citation

```bibtex
@article{zhao2026faster,
  title   = {Faster-WAM: Efficient Inference-Time Future Conditioning for Robust World Action Models},
  author  = {Zhao, Weiheng and Jiang, Haoyi and Shi, Xin and Liu, Liu and Huang, Fan and Su, Zhizhong and Sui, Wei and Wang, Xinggang},
  journal = {arXiv preprint arXiv:2608.04404},
  year    = {2026}
}
```