---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- AIC
configs:
- config_name: default
  data_files: data/*/*.parquet
---

# AIC Robotics Dataset

This dataset was created using [LeRobot](https://github.com/huggingface/lerobot) and is designed for training robotics imitation learning models on **SFP and SC cable plugging tasks** on a modular task board.

## Dataset Overview

- **Robot Type:** AIC
- **Total Episodes:** 80
- **Total Frames:** 22,090
- **Total Tasks:** 12
- **FPS:** 20

## Training Information: Input and Output

To train a model (e.g., using a policy head) with this dataset, the following inputs and outputs should be used:

### 📥 Inputs (Observations)
The model should take the following as input at each timestep:

1.  **Visual Data (RGB Videos):**
    *   `observation.images.left_camera`: (256, 288, 3)
    *   `observation.images.center_camera`: (256, 288, 3)
    *   `observation.images.right_camera`: (256, 288, 3)
2.  **Kinematic State (TCP Offset):**
    *   `observation.tcp_offset.linear`: [x, y, z] values.
    *   `observation.tcp_offset.angular`: [x, y, z] values.
3.  **Task Conditionals (One-hot encoded):**
    *   `plug_type`: [is_sfp, is_sc]
    *   `nic_card`: [is_0, is_1, is_2, is_3, is_4]
    *   `sfp_port`: [is_0, is_1]
    *   `sc_rail`: [is_0, is_1]
    *   `sc_port`: [is_0, is_1, is_2, is_3, is_4]

### 📤 Outputs (Actions)
The model should predict the following values for the next step:

1.  **TCP Offset Delta (Actions):**
    *   `action.tcp_offset.linear`: Target [x, y, z] offset.
    *   `action.tcp_offset.angular`: Target [x, y, z] offset.

---

## Dataset Structure

The core features as defined in `meta/info.json` are:

```json
{
    "observation.images.left_camera": "video (256, 288, 3)",
    "observation.images.center_camera": "video (256, 288, 3)",
    "observation.images.right_camera": "video (256, 288, 3)",
    "observation.tcp_offset.linear.x/y/z": "float32",
    "observation.tcp_offset.angular.x/y/z": "float32",
    "observation.task_id.plug_type": "float32[2]",
    "observation.task_id.nic_card": "float32[5]",
    "observation.task_id.sfp_port": "float32[2]",
    "observation.task_id.sc_rail": "float32[2]",
    "observation.task_id.sc_port": "float32[5]",
    "action.tcp_offset.linear.x/y/z": "float32",
    "action.tcp_offset.angular.x/y/z": "float32"
}
```

## Dataset Files
- `data/`: Contains episode data in Parquet format.
- `meta/`: Contains dataset information, statistics, and task definitions.
- `videos/`: Contains the raw video streams for each camera view.