Episodes Preview SO-101 Visualizer
60 episodes Β· 30 fps Β· 2 cameras Β· 640Γ—480 av1

LeIsaac Pick Orange Dataset

A robot manipulation dataset for learning to pick and place oranges, collected in Isaac Lab simulation environment.

Task Description

The robot performs the following subtasks:

  1. Approach orange - Move gripper towards the orange
  2. Grab orange - Close gripper to grasp the orange
  3. Move to plate - Transport the orange towards the plate
  4. Place orange - Open gripper to release the orange on the plate
  5. Retreat - Move away from the placement location

SARM Annotations

This dataset includes Stage-Aware Reward Modeling (SARM) sparse annotations for all 60 episodes. Each episode is annotated with temporal boundaries for each subtask.

Temporal Proportions (Average)

Subtask Proportion
approach orange 17.4%
grab orange 11.3%
move to plate 17.9%
place orange 20.9%
retreat 32.5%

Annotation Files

  • meta/sarm_annotations_sparse.json - JSON format annotations
  • meta/temporal_proportions_sparse.json - Average temporal proportions
  • Episode-level annotations in parquet files with columns:
    • sparse_subtask_names
    • sparse_subtask_start_times
    • sparse_subtask_end_times
    • sparse_subtask_start_frames
    • sparse_subtask_end_frames

Dataset Structure

leisaac-pick-orange/
β”œβ”€β”€ meta/
β”‚   β”œβ”€β”€ info.json
β”‚   β”œβ”€β”€ stats.json
β”‚   β”œβ”€β”€ tasks.parquet
β”‚   β”œβ”€β”€ episodes/
β”‚   β”‚   └── chunk-000/
β”‚   β”‚       └── file-*.parquet
β”‚   β”œβ”€β”€ sarm_annotations_sparse.json
β”‚   └── temporal_proportions_sparse.json
β”œβ”€β”€ data/
β”‚   └── chunk-000/
β”‚       └── file-*.parquet
└── videos/
    β”œβ”€β”€ observation.images.front/
    β”‚   └── chunk-000/
    β”‚       └── file-*.mp4
    └── observation.images.wrist/
        └── chunk-000/
            └── file-*.mp4

Usage

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

# Load the dataset
dataset = LeRobotDataset("SunyGala/leisaac-pick-orange")

# Access episodes
for i in range(len(dataset)):
    episode = dataset[i]
    # episode contains observation images, robot state, and actions

License

Apache 2.0

Citation

If you use this dataset, please cite the relevant works:

@software{lerobot2024,
  title = {LeRobot: Making AI for Robotics more accessible},
  author = {LeRobot team},
  url = {https://github.com/huggingface/lerobot}
}

Acknowledgments

  • Dataset generated using Isaac Lab
  • Annotations generated using SiliconFlow API with Qwen3-VL model
Downloads last month
82