vla-ustc/libero_10_deluxe

Robot Manipulation Dataset with Trajectory Annotations (Deluxe Edition)

Dataset Description

This enhanced dataset contains robot manipulation demonstrations with:

• Image observations: RGB images from robot camera
• Wrist camera views: End-effector perspective images
• Rich conversations: Task descriptions with semantic subtask annotations
• Trajectory annotations: Visual grounding through trajectory points
• Reverse trajectories: Return-to-initial-position augmented data

Dataset Structure

Fields

• image: Primary camera RGB image
• wrist: Wrist/end-effector camera image
• conversations: Structured task dialogue
  • Task description with semantic subtask decomposition
  • Depth map information
  • Trajectory predictions
  • Action token sequences
• annotation: Trajectory coordinates (null for standard frames, populated for annotated frames)

Data Composition

This dataset combines:

1. Forward task data: Complete task demonstrations with semantic descriptions
2. Reverse trajectory data: Return-to-initial-position augmentation (shifted indices)

Total samples: ~42K+ (21K+ forward + 21K+ reverse)

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("DannyJun/libero_10_deluxe")
train_data = dataset['train']

# Access sample
sample = train_data[0]
print("Task:", sample['conversations']['value'][0])
print("Annotation:", sample['annotation'])

Example Conversation Format

Human: The task is pick up the book and place it in the back compartment of the caddy. What is the action that the robot should take? To figure out the action that the robot should take, let's think through it step by step...

Assistant: The depth map of the first image is .... The number of subtasks is 2. The subtask goals are: pick up the book at [116, 49] and place the book in the back compartment of the caddy at [128, 71]. The trajectory of the end effector in the first image is [[116, 49], [120, 55], [128, 71]]...

Tasks Included

• Object placement and manipulation
• Container operations (microwave, drawer, cabinet)
• Multi-object coordination
• Sequential manipulation tasks
• Reverse trajectory tasks (return to initial position)

Citation

@dataset{robot_manipulation_deluxe,
  title={Robot Manipulation Dataset with Trajectory Annotations (Deluxe Edition)},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/DannyJun/libero_10_deluxe}
}

License

MIT License