THE SOURCE INDEX / 03

Knowledge resources

Find original papers, datasets, code and technical documentation.

DATA / RESEARCH MATERIAL

Start with the data.

Choose data for your task, then check its origin, modalities and terms.

Entries link to original access instructions. Collections can overlap with their constituent datasets; general anomaly benchmarks do not replace damage validation on actual equipment.

01
Public dataset2022

Touch and Go

University of Michigan / CMU

Human-collected egocentric video and GelSight readings from touching objects in natural environments.

Data origin
Measured · human collection
Modalities
Vision, touch, material labels

CC BY 4.0; download from the official page.

02
Public dataset2019

VisGel

MIT CSAIL

Synchronised visual and tactile sequences of robot contact with everyday objects, for cross-modal prediction.

Data origin
Measured · robot collection
Modalities
RGB, GelSight touch

See the author page and repository for access; verify terms in the original files.

03
Public dataset2023

ObjectFolder-Real

Stanford Vision and Learning Lab and collaborators

Meshes, videos, impact sounds and tactile measurements for 100 real objects.

Data origin
Measured · multisensory
Modalities
Vision, audio, touch, meshes

Use the official ObjectFolder-Real download; follow the source terms.

04
Public dataset2022

ObjectFolder 2.0

Stanford Vision and Learning Lab and collaborators

Implicit representations of 1,000 objects queried by viewpoint, impact location and contact conditions.

Data origin
Synthetic · neural representations
Modalities
Vision, audio, touch

Download and query the models following project instructions; distinguish them from measured touch.

05
Public dataset2023–2024

Open X-Embodiment

Open X-Embodiment Collaboration

A collection of robot-manipulation datasets with a common format, dataset inventory and RT-X research.

Data origin
Collection · multiple constituent datasets
Modalities
Images, states, actions; subset-dependent

Terms vary by constituent dataset. Separate BridgeData and RH20T entries provide details, not additional independent volume.

06
Public dataset2024

DROID

DROID Dataset Team

Multi-scene Franka manipulation trajectories, with language annotations, camera calibration and a data visualiser.

Data origin
Measured · teleoperated demonstrations
Modalities
Multi-view images, states, actions, language

The official page supplies Colab and cloud-data paths; check terms for data, code and supplementary annotations.

07
Public dataset2023

BridgeData V2

UC Berkeley RAIL and collaborators

Teleoperated and scripted trajectories spanning pick-and-place, doors, cloth folding and sweeping.

Data origin
Measured · teleoperation and scripted collection
Modalities
Images, actions, language; depth in some subsets

Collection modes have separate downloads; portions also appear in Open X-Embodiment.

08
Public dataset2024

RH20T

Shanghai Jiao Tong University

Contact-rich robot sequences paired with human demonstrations, multi-view vision, force, audio and actions.

Data origin
Measured · multiple embodiments
Modalities
RGB-D, 6-axis force, audio, actions

RH20T-C and RH20T-NC have different licences; fingertip touch is limited to Cfg. 7. Google Drive and Baidu links are provided.

09
Public dataset2023

LIBERO

Lifelong Robot Learning collaborators

Teleoperated demonstrations in controlled simulated tasks for studying transfer of spatial, object and goal knowledge.

Data origin
Simulation · manipulation benchmark
Modalities
Images, states, actions, task descriptions

Download scripts are in the official repository; these are simulated data. See the repository licence.

10
Public dataset2021–

robomimic · Demonstration datasets

Stanford / UT Austin and collaborators

An index of real and simulated manipulation demonstrations, with download, preprocessing and training interfaces.

Data origin
Measured / simulated · task-dependent
Modalities
Images, states, actions; HDF5

This is a dataset index; download each subset and check its source, format and terms.

11
Public dataset2025

PhysicalAI · GR00T X-Embodiment Sim

NVIDIA

NVIDIA’s cross-embodiment simulated manipulation data for GR00T-related training and research.

Data origin
Simulation · cross-embodiment
Modalities
Observations, states, actions; see data card

The data card specifies CC BY 4.0. Use the official files and check embodiment and format for the version.

12
Public dataset2023

AeBAD · Aero-engine Blade Anomaly Detection

Zilong Zhang and collaborators

Single-blade images (AeBAD-S) and blade videos (AeBAD-V) for anomaly detection under changes in illumination, view and scale.

Data origin
Measured · actual blades and defect test specimens
Modalities
RGB images, videos, annotations; subset-dependent

Data: CC BY 4.0. The source repository links Google Drive and Tianyi downloads. Distinguish engineered specimen defects from in-service damage.

13
Public dataset2025

BladeSynth

Mohammed Eltoum and collaborators

Synthetic turbine-blade renders covering normal, dent, nick, scratch and corrosion classes, with defect segmentation masks.

Data origin
Synthetic · blade defects
Modalities
RGB images, classes, segmentation masks

The original record is on Figshare; generation files and scripts are in the author repository. Check the dataset-record licence.

14
Public dataset2019

MVTec AD

MVTec Software

Normal and anomalous industrial object and texture images, with pixel-level defect annotations.

Data origin
Measured · industrial anomalies
Modalities
RGB images, pixel masks

CC BY-NC-SA 4.0; access through the official form. Non-commercial use.

15
Public dataset2025–2026

MVTec AD 2

MVTec Software

More demanding backgrounds and illumination conditions, with public and hidden-annotation test splits.

Data origin
Measured · industrial anomalies
Modalities
Images, public-split test masks

CC BY-NC-SA 4.0; the hidden-annotation split is evaluated by the official server.

16
Public dataset2022

MVTec 3D-AD

MVTec Software

Object-surface scans acquired with an industrial 3D sensor for anomaly detection and localisation.

Data origin
Measured · 3D scans
Modalities
3D geometry, colour, annotations

CC BY-NC-SA 4.0; data and evaluation code are linked from the official page.

17
Public dataset2022

VisA

Amazon Science

Visual anomaly data for PCBs and everyday objects, including scratches, dents and missing components.

Data origin
Measured · visual anomalies
Modalities
RGB images, anomaly labels and masks

Data: CC BY 4.0; code: Apache 2.0. Download and splits are in the official repository.

18
Public dataset2018

SDNET2018

Utah State University

Cracked and non-cracked concrete deck, wall and pavement images, including shadows and surface clutter.

Data origin
Measured · concrete cracks
Modalities
RGB images, image-level labels

CC BY 4.0; image-level classification labels, not pixel-level crack segmentation.

BROWSE BY RESEARCH DIRECTIONResearch directions & topics ↗