README.md

metadata

license: mit
task_categories:
  - visual-question-answering
  - image-classification
language:
  - en
tags:
  - robotics
  - condition-checking
  - multi-modal
  - vision-language
size_categories:
  - n<1K

Condition Checking Dataset

This dataset contains condition checking conversations for robotics applications, with embedded base64 images from multiple camera viewpoints.

Dataset Description

• Task: Visual condition checking (True/False questions about robot states)
• Modality: Multi-modal (text + images)
• Domain: Robotics manipulation tasks
• Format: Conversational format suitable for VLM training

Dataset Structure

Data Fields

• id: Unique identifier for each sample
• images: Dictionary containing base64-encoded images from multiple camera viewpoints
• conversations: List of conversation turns (human question + assistant answer)

Camera Viewpoints

The dataset includes images from 5 camera viewpoints:

• observation_images_chest
• observation_images_left_eye
• observation_images_left_wrist
• observation_images_right_eye
• observation_images_right_wrist

Each sample contains approximately 30 images total (6 per camera).

Sample Structure

{
  "id": "frame_index_position_part",
  "images": {
    "camera_key": ["base64_image_1", "base64_image_2", ...],
    ...
  },
  "conversations": [
    {
      "from": "human",
      "value": "Here are the observations... condition: (object is grasped) ..."
    },
    {
      "from": "gpt", 
      "value": "True"
    }
  ]
}

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("jeffshen4011/condition-checking-dataset")

# Access a sample
sample = dataset["train"][0]
print(f"Question: {sample['conversations'][0]['value'][:100]}...")
print(f"Answer: {sample['conversations'][1]['value']}")
print(f"Number of camera views: {len(sample['images'])}")

Dataset Statistics

• Training samples: 9
• Camera viewpoints: 5
• Images per sample: ~30
• Image format: Base64-encoded PNG
• Task type: Binary classification (True/False)

Applications

This dataset is designed for:

• Training vision-language models for robotics condition checking
• Multi-modal reasoning tasks
• Robot state verification
• Visual question answering in manipulation contexts

Citation

If you use this dataset, please cite:

@dataset{condition_checking_dataset,
  title={Condition Checking Dataset for Robotics},
  author={Research Team},
  year={2025},
  url={https://huggingface.co/datasets/jeffshen4011/condition-checking-dataset}
}