CLARU
s-tianMIT

VP2 Benchmark Data

Training datasets and benchmark data for video prediction in robotic control tasks, corresponding to the ICLR 2023 paper 'A Control-Centric Benchmark for Video Prediction'.

Downloads190

Technical Profile

Modalities
rgb
Environment
simulation
Task Types
video prediction
License
MIT
Part of the VP2 Benchmark Data family

Access

View on HuggingFace

Need custom rgb data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets

UR Ultrasound Robotics Dataset

A robotics dataset collected using LeRobot with a UR robot performing ultrasound-related tasks, featuring multi-view RGB imagery including base, wrist, and ultrasound camera feeds.

rgbproprioception
null downloadsJul 2026apache-2.0

CSI-Agent Towel Folding Dataset

A dataset of towel folding demonstrations collected using LeRobot with a bimanual SO-101 robot. Contains 50 episodes of towel folding tasks with multi-camera RGB observations and joint position actions.

rgbproprioception
null downloadsJul 2026apache-2.0

WGO-Bench Localization Given Labels

A temporal localization benchmark where models are given robot manipulation videos and shuffled event labels with multiplicities, and must predict the start and end times for each labeled event occurrence. Derived from Macrodata Labs' WGO-Bench with 100 episodes split into 80 training and 20 test episodes.

rgblanguage
null downloadsJul 2026cc-by-nc-sa-4.0

Egocentric Residential Video: 100,000 Hours of First-Person Household Tasks

A 100,000-hour collection of high-quality first-person videos recorded in real residential environments across Brazil at 1080p resolution or higher, capturing natural household chores with structured annotations for tasks, actions, objects, and hand-object interactions.

rgb
null downloadsJul 2026datoric-commercial

Egocentric Residential Video: 100,000 Hours of First-Person Household Tasks

A 100,000-hour collection of 1080p+ first-person videos of household chores recorded in real residential environments across Brazil, with structured annotations for tasks, actions, objects, and hand-object interactions. Built for physical AI, robotics, egocentric vision, and embodied AI research focusing on non-cooking residential tasks.

rgb
null downloadsJul 2026datoric-commercial

pick_place9_20260709_141045

A LeRobot dataset containing pick and place manipulation tasks collected with a SO Follower robot platform at 30 fps.

rgbproprioception
null downloadsJul 2026apache-2.0