---
# Example metadata to be added to a dataset card.  
# Full dataset card template at https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/datasetcard_template.md
language:
- en
license: mit  # Example: apache-2.0 or any license from https://hf.co/docs/hub/repositories-licenses
tags:
- robotics
- manipulation
- rearrangement
- computer-vision
- reinforcement-learning
- imitation-learning
- rgbd
- rgb
- depth
- low-level-control
- whole-body-control
- home-assistant
- simulation
- maniskill
annotations_creators:
- machine-generated  # Generated from RL policies with filtering
language_creators:
- machine-generated
language_details: en-US
pretty_name: ManiSkill-HAB SetTabkle Dataset
size_categories:
- 1M<n<10M  # Dataset has 8K episodes with 1.6M transitions
# source_datasets:  # None, original
task_categories:
- robotics
- reinforcement-learning
task_ids:
- grasping
- task-planning

configs:
  - config_name: pick-013_apple
    data_files:
    - split: trajectories
      path: pick/013_apple.h5
    - split: metadata
      path: pick/013_apple.json
  
  - config_name: pick-024_bowl
    data_files:
    - split: trajectories
      path: pick/024_bowl.h5
    - split: metadata
      path: pick/024_bowl.json

  - config_name: place-013_apple
    data_files:
    - split: trajectories
      path: place/013_apple.h5
    - split: metadata
      path: place/013_apple.json
  
  - config_name: place-024_bowl
    data_files:
    - split: trajectories
      path: place/024_bowl.h5
    - split: metadata
      path: place/024_bowl.json

  - config_name: open-fridge
    data_files:
    - split: trajectories
      path: open/fridge.h5
    - split: metadata
      path: open/fridge.json

  - config_name: open-kitchen_counter
    data_files:
    - split: trajectories
      path: open/kitchen_counter.h5
    - split: metadata
      path: open/kitchen_counter.json

  - config_name: close-fridge
    data_files:
    - split: trajectories
      path: close/fridge.h5
    - split: metadata
      path: close/fridge.json
  
  - config_name: close-kitchen_counter
    data_files:
    - split: trajectories
      path: close/kitchen_counter.h5
    - split: metadata
      path: close/kitchen_counter.json

# # Optional. This part can be used to store the feature types and size of the dataset to be used in python. This can be automatically generated using the datasets-cli.
# dataset_info:
#   features:
#     - name: {feature_name_0}    # Example: id
#       dtype: {feature_dtype_0}  # Example: int32
#     - name: {feature_name_1}    # Example: text
#       dtype: {feature_dtype_1}  # Example: string
#     - name: {feature_name_2}    # Example: image
#       dtype: {feature_dtype_2}  # Example: image
#     # Example for SQuAD:
#     # - name: id
#     #   dtype: string
#     # - name: title
#     #   dtype: string
#     # - name: context
#     #   dtype: string
#     # - name: question
#     #   dtype: string
#     # - name: answers
#     #   sequence:
#     #     - name: text
#     #       dtype: string
#     #     - name: answer_start
#     #       dtype: int32
#   config_name: {config_name}  # Name of the dataset subset. Example for glue: sst2
#   splits:
#     - name: {split_name_0}                  # Example: train
#       num_bytes: {split_num_bytes_0}        # Example for SQuAD: 79317110
#       num_examples: {split_num_examples_0}  # Example for SQuAD: 87599
#   download_size: {dataset_download_size}   # Example for SQuAD: 35142551
#   dataset_size: {dataset_size}             # Example for SQuAD: 89789763

# It can also be a list of multiple subsets (also called "configurations"):
# ```yaml
# dataset_info:
#   - config_name: {config0}
#     features:
#       ...
#   - config_name: {config1}
#     features:
#       ...
# ```

# # Optional. If you want your dataset to be protected behind a gate that users have to accept to access the dataset. More info at https://huggingface.co/docs/hub/datasets-gated
# extra_gated_fields:
# - {field_name_0}: {field_type_0}  # Example: Name: text
# - {field_name_1}: {field_type_1}  # Example: Affiliation: text
# - {field_name_2}: {field_type_2}  # Example: Email: text
# - {field_name_3}: {field_type_3}  # Example for speech datasets: I agree to not attempt to determine the identity of speakers in this dataset: checkbox
# extra_gated_prompt: {extra_gated_prompt}  # Example for speech datasets: By clicking on “Access repository” below, you also agree to not attempt to determine the identity of speakers in the dataset.

# # Optional. Add this if you want to encode a train and evaluation info in a structured way for AutoTrain or Evaluation on the Hub
# train-eval-index:
#   - config: {config_name}           # The dataset subset name to use. Example for datasets without subsets: default. Example for glue: sst2
#     task: {task_name}               # The task category name (same as task_category). Example: question-answering
#     task_id: {task_type}            # The AutoTrain task id. Example: extractive_question_answering
#     splits:
#       train_split: train            # The split to use for training. Example: train
#       eval_split: validation        # The split to use for evaluation. Example: test
#     col_mapping:                    # The columns mapping needed to configure the task_id.
#     # Example for extractive_question_answering:
#       # question: question
#       # context: context
#       # answers:
#       #   text: text
#       #   answer_start: answer_start
#     metrics:
#       - type: {metric_type}         # The metric id. Example: wer. Use metric id from https://hf.co/metrics
#         name: {metric_name}         # Tne metric name to be displayed. Example: Test WER
---

# ManiSkill-HAB SetTable Dataset

**[Paper](https://arxiv.org/abs/2412.13211)** 
| **[Website](https://arth-shukla.github.io/mshab)** 
| **[Code](https://github.com/arth-shukla/mshab)** 
| **[Models](https://huggingface.co/arth-shukla/mshab_checkpoints)** 
| **[(Full) Dataset](https://arth-shukla.github.io/mshab/#dataset-section)** 
| **[Supplementary](https://sites.google.com/view/maniskill-hab)**


<!-- Provide a quick summary of the dataset. -->

Whole-body, low-level control/manipulation demonstration dataset for ManiSkill-HAB SetTable.

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

Demonstration dataset for ManiSkill-HAB SetTable. Each subtask/object combination (e.g pick 013_apple) has 1000 successful episodes (200 samples/demonstration) gathered using [RL policies](https://huggingface.co/arth-shukla/mshab_checkpoints) fitered for safe robot behavior with a rule-based event labeling system.

SetTable contains the Pick, Place, Open, and Close subtasks. Relative to the other MS-HAB long-horizon tasks (TidyHouse, PrepareGroceries), SetTable Pick, Place, Open, and Close are easy difficulty (on a scale of easy-medium-hard). The difficulty of SetTable primarily comes from skill chaining rather than individual subtasks.

### Related Datasets

Full information about the MS-HAB datasets (size, difficulty, links, etc), including the other long horizon tasks, are available [on the ManiSkill-HAB website](https://arth-shukla.github.io/mshab/#dataset-section).

- [ManiSkill-HAB TidyHouse Dataset](https://huggingface.co/datasets/arth-shukla/MS-HAB-TidyHouse)
- [ManiSkill-HAB PrepareGroceries Dataset](https://huggingface.co/datasets/arth-shukla/MS-HAB-PrepareGroceries)

## Uses

<!-- Address questions around how the dataset is intended to be used. -->

### Direct Use

This dataset can be used to train vision-based learning from demonstrations and imitation learning methods, which can be evaluated with the [MS-HAB environments](https://github.com/arth-shukla/mshab). This dataset may be useful as synthetic data for computer vision tasks as well.

### Out-of-Scope Use

While blind state-based policies can be trained on this dataset, it is recommended to train vision-based policies to handle collisions and obstructions.

## Dataset Structure

Each subtask/object combination has files `[SUBTASK]/[OBJECT].json` and `[SUBTASK]/[OBJECT].h5`. The JSON file contains episode metadata, event labels, etc, while the HDF5 file contains the demonstration data.

## Dataset Creation

<!-- TODO (arth): link paper appendix, maybe html, for the event labeling system -->
The data is gathered using [RL policies](https://huggingface.co/arth-shukla/mshab_checkpoints) fitered for safe robot behavior with a rule-based event labeling system.

## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

The dataset is purely synthetic.

While MS-HAB supports high-quality ray-traced rendering, this dataset uses ManiSkill's default rendering for data generation due to efficiency. However, users can generate their own data with the [data generation code](https://github.com/arth-shukla/mshab/blob/main/mshab/utils/gen/gen_data.py).

<!-- TODO (arth): citation -->
## Citation

```
@inproceedings{shukla2025maniskillhab,
  author       = {Arth Shukla and
                  Stone Tao and
                  Hao Su},
  title        = {ManiSkill-HAB: {A} Benchmark for Low-Level Manipulation in Home Rearrangement
                  Tasks},
  booktitle    = {The Thirteenth International Conference on Learning Representations,
                  {ICLR} 2025, Singapore, April 24-28, 2025},
  publisher    = {OpenReview.net},
  year         = {2025},
  url          = {https://openreview.net/forum?id=6bKEWevgSd},
  timestamp    = {Thu, 15 May 2025 17:19:05 +0200},
  biburl       = {https://dblp.org/rec/conf/iclr/ShuklaTS25.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}
```