Instructions to use huangrm/MINT-libero with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use huangrm/MINT-libero with LeRobot:
- Notebooks
- Google Colab
- Kaggle
Add model card for MINT-libero
#1
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
library_name: lerobot
|
| 4 |
+
pipeline_tag: robotics
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# MINT: Mimic Intent, Not Just Trajectories
|
| 8 |
+
|
| 9 |
+
MINT (Mimic Intent, Not just Trajectories) is a framework for end-to-end imitation learning in dexterous manipulation. It explicitly disentangles behavior intent from execution details by learning a hierarchical, multi-scale token representation of actions.
|
| 10 |
+
|
| 11 |
+
- **Paper:** [Mimic Intent, Not Just Trajectories](https://huggingface.co/papers/2602.08602)
|
| 12 |
+
- **Project Page:** [https://renming-huang.github.io/MINT/](https://renming-huang.github.io/MINT/)
|
| 13 |
+
- **GitHub Repository:** [https://github.com/RenMing-Huang/MINT](https://github.com/RenMing-Huang/MINT)
|
| 14 |
+
|
| 15 |
+
## Model Description
|
| 16 |
+
|
| 17 |
+
While imitation learning has achieved success in robotic manipulation, models often struggle with adaptation and transfer because they mimic raw trajectories. MINT addresses this by disentangling behavior intent from execution details through multi-scale frequency-space tokenization.
|
| 18 |
+
- **Intent Tokens:** Capture low-frequency global structure to facilitate planning and transfer.
|
| 19 |
+
- **Execution Tokens:** Encode high-frequency details to enable precise adaptation to environmental dynamics.
|
| 20 |
+
- **Autoregressive Reasoning:** The policy generates trajectories through next-scale autoregression, performing progressive intent-to-execution reasoning.
|
| 21 |
+
|
| 22 |
+
## Usage
|
| 23 |
+
|
| 24 |
+
This model is designed to be used with the [LeRobot](https://github.com/huggingface/lerobot) library.
|
| 25 |
+
|
| 26 |
+
### Evaluation
|
| 27 |
+
|
| 28 |
+
To evaluate the policy in the LIBERO environment, use the following command (requires the [MINT-tokenizer-libero](https://huggingface.co/huangrm/MINT-tokenizer-libero)):
|
| 29 |
+
|
| 30 |
+
```bash
|
| 31 |
+
lerobot-eval \
|
| 32 |
+
--policy.path=huangrm/MINT-libero \
|
| 33 |
+
--policy.vqvae_name_or_path=huangrm/MINT-tokenizer-libero \
|
| 34 |
+
--env.type=libero \
|
| 35 |
+
--env.task=libero_10,libero_object,libero_spatial,libero_goal \
|
| 36 |
+
--eval.batch_size=1 \
|
| 37 |
+
--eval.n_episodes=2 \
|
| 38 |
+
--seed=42 \
|
| 39 |
+
--policy.n_action_steps=4
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
## Citation
|
| 43 |
+
|
| 44 |
+
```bibtex
|
| 45 |
+
@article{huang2026mimic,
|
| 46 |
+
title={Mimic Intent, Not Just Trajectories},
|
| 47 |
+
author={Huang, Renming and Zeng, Chendong and Tang, Wenjing and Cai, Jintian and Lu, Cewu and Cai, Panpan},
|
| 48 |
+
journal={arXiv preprint arXiv:2602.08602},
|
| 49 |
+
year={2026}
|
| 50 |
+
}
|
| 51 |
+
```
|