datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
libero_safetySafe_and_Unsafe_Behaviours
Dataset Card for safe_unsafe_behaviours
This is a FiftyOne dataset with 691 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/Safe_and_Unsafe_Behaviours")
# Launch the App
session = fo.launch_app(dataset)
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/Safe_and_Unsafe_Behaviours.safewatch-genai
SafeWatch-TV2V Dataset
Overview
This dataset combines SafeWatch-Bench original videos and TV2V generated videos for video safety detection and robustness testing.
Dataset Structure
Three Main Components:
TV2V-output: Generated videos with prompts
Videos: 1,040 files
Metadata: 1,213 entries in TV2V-output.jsonl
Content: Generated videos with their corresponding prompts
TV2V-input: Original videos used for generation
Videos: 410 files… See the full description on the dataset page: https://huggingface.co/datasets/EchoRaven/safewatch-genai.safer-ar-dataput_money_in_safe_25_08_21_lerobotv2.1Safe_Unsafe_Test-Understanding-text-qwen-embs
Dataset Card for pjramg/Safe_Unsafe_Test
This is a FiftyOne dataset with 40 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("pjramg/Safe_Unsafe_Test-Understanding-text-qwen-embs")
# Launch the App
session = fo.launch_app(dataset)… See the full description on the dataset page: https://huggingface.co/datasets/pjramg/Safe_Unsafe_Test-Understanding-text-qwen-embs.cubes_push_safe_clean_100rewind_piper_demo_shift_k4_safeSafeVLA-Bench-Dataset
SafeVLA PickCube dataset card
Code, converters, split audit, and benchmark reports: https://github.com/Jatshi/SafeVLA-Bench.
Contents
pickcube_state_h16_full1000_holdout_v1.npz: 62,681 state/action-chunk samples from 1,000 CPU replays after excluding the five evaluation-seed episodes.
pickcube_state_h16_v1.npz: 6,220 state/action-chunk samples from the first 100 CPU replays; retained only for diagnostic comparison, not the final result.… See the full description on the dataset page: https://huggingface.co/datasets/jatshi/SafeVLA-Bench-Dataset.Safe_Unsafe_Test-Understanding
Dataset Card for pjramg/Safe_Unsafe_Test
This is a FiftyOne dataset with 40 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("pjramg/Safe_Unsafe_Test-Understanding")
# Launch the App
session = fo.launch_app(dataset)… See the full description on the dataset page: https://huggingface.co/datasets/pjramg/Safe_Unsafe_Test-Understanding.SafeCritCR
SafeCritCR
This dataset consists of 1,200 videos of CommonRoad scenarios that include 8 different types of safety criticalities.
They have been rendered using the Frenetix Motion Planner.
The complete set of samples follows a longtail distribution in terms of their labels. We constructed a uniformly distributed subset with 5 classes and 300 entries.
Videos in this subset are tagged with "in_balanced_subset": true.
Videos
The MP4s depict the scenarios from a bird's… See the full description on the dataset page: https://huggingface.co/datasets/Pusch/SafeCritCR.eval_right_pick_place_v6_minmax_safetyThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"names": [
"arm_left_shoulder_pan.pos",
"arm_left_shoulder_lift.pos",
"arm_left_elbow_flex.pos",
"arm_left_wrist_flex.pos",
"arm_left_wrist_yaw.pos",
"arm_left_wrist_roll.pos"… See the full description on the dataset page: https://huggingface.co/datasets/yj678/eval_right_pick_place_v6_minmax_safety.cubes_push_bottom_to_top_safeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "arx5",
"total_episodes": 10,
"total_frames": 3572,
"total_tasks": 1,
"total_videos": 20,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 20,
"splits": {
"train": "0:10"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/pravsels/cubes_push_bottom_to_top_safe.vlsa-dual-safety-eval
VLSA Dual-Arm Safety Evaluation (AEGIS, dual-arm extension)
Quantitative evaluation of a plug-and-play CBF safety layer (VLSA/AEGIS) extended to a
dual-arm setting, where two Franka Panda arms face each other across a table and each treats
the opposing arm's end-effector as a dynamic ellipsoidal obstacle.
This is a dual-arm extension of VLSA: Vision-Language-Action Models with Plug-and-Play Safety
Constraint Layer (AEGIS, arXiv:2512.11891). The original work is single-arm with… See the full description on the dataset page: https://huggingface.co/datasets/morealcholplz/vlsa-dual-safety-eval.cubes_push_right_to_left_safeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "arx5",
"total_episodes": 10,
"total_frames": 3686,
"total_tasks": 1,
"total_videos": 20,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 20,
"splits": {
"train": "0:10"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/pravsels/cubes_push_right_to_left_safe.MCV_SafetyBenchAgarBadge_v1_4th_Thermal-Safety-Logic_42_20251203
AgarBadge_v1_4th_Thermal-Safety-Logic_42_20251203
This dataset was created using LeRobot.
YFE-v1-30-6rd-260107-safetyeval_YFE-final-v5-gt-action-safetyThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 10,
"total_frames": 12805,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:10"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/lerobotForScienceEdu/eval_YFE-final-v5-gt-action-safety.libero_safety_v1
LIBERO Safety
This repository contains 50-scene synthetic LIBERO safety validation sets.
v1: original LIBERO safety image set.
v2: regenerated set using the latest scene configuration and saved manual positions.
v3: same scenes and labels as v2, with a static non-colliding MuJoCo white cutting board fixture added on the table; original V2 object and fixture states are unchanged.
v4: corrected zoomed non-reference set with pickable target objects for video rollouts.
v5: corrected… See the full description on the dataset page: https://huggingface.co/datasets/saaduddinM/libero_safety_v1.Safe_and_Unsafe_Behaviours
Dataset Card for safe_unsafe_behaviours
This is a FiftyOne dataset with 691 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/Safe_and_Unsafe_Behaviours")
# Launch the App
session = fo.launch_app(dataset)
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/yangyayaa/Safe_and_Unsafe_Behaviours.YFE-v2-130-safety-mergedThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 130,
"total_frames": 231377,
"total_tasks": 2,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:130"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/lerobotForScienceEdu/YFE-v2-130-safety-merged.AgarBadge_v1_4th_Thermal-Safety-Logic_91_20251203
AgarBadge_v1_4th_Thermal-Safety-Logic_91_20251203
This dataset was created using LeRobot.
Rock_Acid-129-safetyThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 133,
"total_frames": 188905,
"total_tasks": 2,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:133"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/lerobotForScienceEdu/Rock_Acid-129-safety.Safe_Unsafe_Test-Understanding-output-labels-qwen
Dataset Card for Safe-Unsafe-Video_Understanding
This is a FiftyOne dataset with 40 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("pjramg/Safe_Unsafe_Test-Understanding-output-labels-qwen")
# Launch the App
session =… See the full description on the dataset page: https://huggingface.co/datasets/pjramg/Safe_Unsafe_Test-Understanding-output-labels-qwen.Rock_Acid-130-v3-safetey-merged-251231This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 134,
"total_frames": 190144,
"total_tasks": 2,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:134"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/lerobotForScienceEdu/Rock_Acid-130-v3-safetey-merged-251231.cubes_push_left_to_right_safeThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v2.1",
"robot_type": "arx5",
"total_episodes": 10,
"total_frames": 3237,
"total_tasks": 1,
"total_videos": 20,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 20,
"splits": {
"train": "0:10"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/pravsels/cubes_push_left_to_right_safe.eval_safety-testThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 1,
"total_frames": 550,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 10,
"splits": {
"train": "0:1"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/irvinh/eval_safety-test.Rock_Acid-130-v3-safetey-mergedThis dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 134,
"total_frames": 190144,
"total_tasks": 2,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:134"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/lerobotForScienceEdu/Rock_Acid-130-v3-safetey-merged.Rock_Acid-v4-30-6th-safety-260110This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "so101_follower",
"total_episodes": 30,
"total_frames": 52208,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:30"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/lerobotForScienceEdu/Rock_Acid-v4-30-6th-safety-260110.