# In-Context Imitation Learning via Next-Token Prediction
by Max (Letian) Fu*, Huang Huang*, Gaurav Datta*, Lawrence Yunliang Chen, William Chung-Ho Panitch, Fangchen Liu, Hui Li, and Ken Goldberg at UC Berkeley and Autodesk (*equal contribution).
[[Paper](https://icrt.dev/files/icrt.pdf)] | [[Project Page](https://icrt.dev/)] | [[Checkpoints](https://huggingface.co/mlfu7/ICRT)] | [[Dataset](https://huggingface.co/datasets/Ravenh97/ICRT-MT)] | [[Citation](#citation)]
This repo contains the checkpoints for *In-Context Imitation Learning via Next-Token Prediction*. We investigate how to bring few-shot, in-context learning capability that exists in next-token prediction models (i.e. GPT) into real-robot imitation learning policies.
In particular, we store the pre-trained vision encoder and ICRT model separately. Please find them in [encoder](crossmae_rtx/cross-mae-rtx-vitb.pth), [ICRT](icrt_vitb_droid_pretrained/icrt_vitb_droid_pretrained.pth), and [ICRT-Llama7B](icrt_llama7b_lora/icrt_llama7b_lora.pth).
Please refer to the [code](https://github.com/Max-Fu/icrt) on installing the repo, training and inferencing the model.
## Dataset Structure
```
ICRT-MT
├── merged_data_part1.hdf5
│ ├── episode_1
│ │ ├── observation
│ │ ├── exterior_image_1_left
│ │ └── exterior_image_2_left
│ │ └── wrist_image_left
│ │ └── cartesian_position
│ │ └── gripper_position
│ │ └── joint_position
│ │ ├── action
│ │ ├── cartesian_velocity
│ │ └── gripper_velocity
│ │ └── joint_velocity
│ │ └── cartesian_position
│ │ └── gripper_position
│ │ └── joint_position
│ │ ├── language_instruction
│ │ ├── language_instruction_2
│ │ ├── language_instruction_3
│ │ ├── language_embedding
│ │ ├── language_embedding_2
│ │ ├── language_embedding_3
│ │ ...
│ ├── episode_2
│ │ ...
│ └── episode_3
│ ...
└── merged_data_part1_keys.json
...
```
## Citation
Please give us a star 🌟 on Github to support us!
Please cite our work if you find our work inspiring or use our code in your work:
```
@article{fu2024icrt,
title={In-Context Imitation Learning via Next-Token Prediction},
author={Letian Fu and Huang Huang and Gaurav Datta and Lawrence Yunliang Chen and William Chung-Ho Panitch and Fangchen Liu and Hui Li and Ken Goldberg},
journal={arXiv preprint arXiv:2408.15980},
year={2024}
}
```