IROS-2025-Challenge-Manip

# IROS-2025-Challenge-Manip
# Dataset Summary 📖

This dataset contains the **IROS Challenge - Manipulation Track** benchmark, organized into **pretrain**, **train**, and **validation** splits.

* **Pretrain split**: \~20,000 single pick-and-place trajectories, packaged into tar files (each containing \~1,000 trajectories).
* **Train split**: task-specific demonstrations, with \~100 trajectories provided per task.
* **Validation split**: includes the test-time scenes and object assets in **USD format**.

Each trajectory in the pretrain and train splits contains:

* **Multi-view video** recordings (three perspectives: head-mounted camera and two wrist cameras)
* **Robot states** (joint positions, gripper states, etc.)
* **Actions** corresponding to the task execution

This dataset is designed to support **pretraining, task-specific fine-tuning, and evaluation** for robotic manipulation in the IROS Challenge setting.

# Get started 🔥
## Download the Dataset
To download the full dataset, you can use the following code. If you encounter any issues, please refer to the official Hugging Face documentation.
```python
from huggingface_hub import snapshot_download

dataset_path = snapshot_download("InternRobotics/IROS-2025-Challenge-Manip", repo_type="dataset")
```

Please execute this Python file to post-process the validation set.
```bash
cd IROS-2025-Challenge-Manip
python dataset_post_processing.py validation
````

## Unzip the pretrain dataset

```bash
cd pretrain
for i in {1..20}; do
echo "Extracting $i.tar.gz ..."
tar -xzf "$i.tar.gz"
done
```

## Dataset Structure

### pretrain Folder hierarchy
```
pretrain
├── 1.tar.gz
│ └── 1/
│ ├── data/
│ ├── meta/
│ └── videos/
├── 2.tar.gz
│ └── 2/
│ ├── data/
│ ├── meta/
│ └── videos/
...
├── 20.tar.gz
└── 20/
├── data/
├── meta/
└── videos/

```

### train Folder hierarchy
```
train
├── collect_three_glues
│   ├── data/
│   ├── meta/
│   └── videos/
├── collect_two_alarm_clocks/
├── collect_two_shoes/
├── gather_three_teaboxes/
├── make_sandwich/
├── oil_painting_recognition/
├── organize_colorful_cups/
├── purchase_gift_box/
├── put_drink_on_basket/
└── sort_waste/

```

### validation Folder hierarchy
```
validation
├── IROS_C_V3_Aloha_seen
│   ├── collect_three_glues
│   │   ├── 000
│   │   │   ├── meta_info.pkl
│   │   │   ├── scene.usd
│   │   │   └── SubUSDs -> ../SubUSDs
│   │   ├── 001/
│   │   ├── 002/
│   │   ├── 003/
│   │   ├── 004/
│   │   ├── 005/
│   │   ├── 006/
│   │   ├── 007/
│   │   ├── 008/
│   │   ├── 009/
│   │   └── SubUSDs
│   │   ├── materials/
│   │   └── textures/
│   ├── collect_two_alarm_clocks/
│   ├── collect_two_shoes/
│   ├── gather_three_teaboxes/
│   ├── make_sandwich/
│   ├── oil_painting_recognition/
│   ├── organize_colorful_cups/
│   ├── purchase_gift_box/
│   ├── put_drink_on_basket/
│   └── sort_waste/
└── IROS_C_V3_Aloha_unseen
├── collect_three_glues/
├── collect_two_alarm_clocks/
├── collect_two_shoes/
├── gather_three_teaboxes/
├── make_sandwich/
├── oil_painting_recognition/
├── organize_colorful_cups/
├── purchase_gift_box/
├── put_drink_on_basket/
└── sort_waste/

```

# License and Citation
All the data and code within this repo are under [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/). Please consider citing our project if it helps your research.

```BibTeX
@misc{contributors2025internroboticsrepo,
title={IROS-2025-Challenge-Manip Colosseum},
author={IROS-2025-Challenge-Manip Colosseum contributors},
howpublished={\url{https://github.com/internrobotics/IROS-2025-Challenge-Manip}},
year={2025}
}
```

# IROS-2025-Challenge-Manip
## 数据集摘要 📖

本数据集为**IROS挑战赛-机械操作赛道**基准数据集，划分为**预训练（pretrain）**、**训练（train）**与**验证（validation）**三个子集。
* **预训练（pretrain）子集**：包含约20000条单次抓取放置轨迹，打包为tar文件（每个文件内含约1000条轨迹）。
* **训练（train）子集**：为面向特定任务的演示数据，每个任务提供约100条轨迹。
* **验证（validation）子集**：包含测试阶段的场景与物体资源，格式为**USD（USD format）**。

预训练与训练子集的每条轨迹均包含以下内容：
* **多视角视频（Multi-view video）**录制内容（三种视角：头戴式相机与两台腕部相机）
* **机器人状态（Robot states）**（关节位置、夹爪状态等）
* **任务执行对应动作（Actions）**

本数据集旨在为IROS挑战赛场景下的机器人机械操作任务，提供**预训练、特定任务微调与模型评估**的支持。

## 快速上手 🔥
### 数据集下载
可通过如下代码下载完整数据集。若遇到问题，请参考Hugging Face官方文档。
python
from huggingface_hub import snapshot_download

dataset_path = snapshot_download("InternRobotics/IROS-2025-Challenge-Manip", repo_type="dataset")

请执行下述脚本完成验证集的后处理：
bash
cd IROS-2025-Challenge-Manip
python dataset_post_processing.py validation

### 解压预训练数据集
bash
cd pretrain
for i in {1..20}; do
echo "Extracting $i.tar.gz ..."
tar -xzf "$i.tar.gz"
done

### 数据集组织结构
#### 预训练（pretrain）文件夹层级

pretrain
├── 1.tar.gz
│ └── 1/
│ ├── data/
│ ├── meta/
│ └── videos/
├── 2.tar.gz
│ └── 2/
│ ├── data/
│ ├── meta/
│ └── videos/
...
├── 20.tar.gz
└── 20/
├── data/
├── meta/
└── videos/

#### 训练（train）文件夹层级

train
├── collect_three_glues
│ ├── data/
│ ├── meta/
│ └── videos/
├── collect_two_alarm_clocks/
├── collect_two_shoes/
├── gather_three_teaboxes/
├── make_sandwich/
├── oil_painting_recognition/
├── organize_colorful_cups/
├── purchase_gift_box/
├── put_drink_on_basket/
└── sort_waste/

#### 验证（validation）文件夹层级

validation
├── IROS_C_V3_Aloha_seen
│ ├── collect_three_glues
│ │ ├── 000
│ │ │ ├── meta_info.pkl
│ │ │ ├── scene.usd
│ │ │ └── SubUSDs -> ../SubUSDs
│ │ ├── 001/
│ │ ├── 002/
│ │ ├── 003/
│ │ ├── 004/
│ │ ├── 005/
│ │ ├── 006/
│ │ ├── 007/
│ │ ├── 008/
│ │ ├── 009/
│ │ └── SubUSDs
│ │ ├── materials/
│ │ └── textures/
│ ├── collect_two_alarm_clocks/
│ ├── collect_two_shoes/
│ ├── gather_three_teaboxes/
│ ├── make_sandwich/
│ ├── oil_painting_recognition/
│ ├── organize_colorful_cups/
│ ├── purchase_gift_box/
│ ├── put_drink_on_basket/
│ └── sort_waste/
└── IROS_C_V3_Aloha_unseen
├── collect_three_glues/
├── collect_two_alarm_clocks/
├── collect_two_shoes/
├── gather_three_teaboxes/
├── make_sandwich/
├── oil_painting_recognition/
├── organize_colorful_cups/
├── purchase_gift_box/
├── put_drink_on_basket/
└── sort_waste/

## 许可与引用
本仓库内的所有数据与代码均遵循[CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/)许可协议。若本数据集对你的研究有所帮助，请考虑引用本项目。

BibTeX
@misc{contributors2025internroboticsrepo,
title={IROS-2025-Challenge-Manip Colosseum},
author={IROS-2025-Challenge-Manip Colosseum contributors},
howpublished={url{https://github.com/internrobotics/IROS-2025-Challenge-Manip}},
year={2025}
}