Models and Datasets

Live leaderboard comparing vision-language-action models on the LIBERO simulation benchmark.

Benchmark that reproduces real robot manipulation tests (RT-1, Octo, OpenVLA) safely in simulation.

Independent leaderboard scoring different open vision-language-action robot models against each other.

Curated GitHub list linking to dozens of useful robotics and computer vision datasets.

Searchable list of every dataset on Hugging Face tagged as robotics, from tiny demos to huge collections.

IEEE's dataset hub filtered to robotics, hosting sensor logs and experiment data from many labs.

Kaggle's search results for community uploaded robotics datasets, good for beginner projects.

Filtered list of datasets specifically used for benchmarking robot navigation methods.

Directory linking robotics research papers to their code and benchmark datasets in one place.

Community collection on Zenodo where robotics researchers share free datasets and code archives.

Live demo hub showcasing community built robot projects from the LeRobot worldwide hackathon.

Official Hugging Face Discord server with active channels where LeRobot users ask and answer questions.

Guide to teaching a real robot arm new skills by recording demonstrations and training with LeRobot.

Quick reference of common LeRobot commands for training, recording data, and running robots.

Explains why community collected robot datasets could become the 'ImageNet' of robotics.

Web app for browsing and viewing robot episodes stored in the LeRobot dataset format.

Official Hugging Face page hosting LeRobot's pretrained models and community robot datasets.

Step by step guide to install LeRobot on your computer so you can start training robot models.

Collection of runnable notebooks that teach LeRobot basics like datasets, training, and evaluation.

Setup guide for the SO-101, a low cost robot arm popular for learning LeRobot hands on.

Fork of the LIBERO simulation benchmark adapted to work with Hugging Face LeRobot pipelines.

Hugging Face org for the global LeRobot hackathon, with community made robot models and demos.

Technical guide to the newest LeRobot dataset format for storing large scale robot trajectories.

Interactive Hugging Face Space that teaches the basics of robot learning step by step.

Case study Space showing how LeRobot took a shirt folding robot from data collection to deployment.

Large scale humanoid robot manipulation dataset from AgiBot, covering many real world tasks.

Code and tools for the AgiBot World humanoid manipulation dataset and benchmark platform.

Code to download, process, and visualize the ARCTIC bimanual hand-object dataset.

Dataset of two hands manipulating objects together, used to study dexterous bimanual manipulation.

Hugging Face mirror of BEHAVIOR-1K household activity data for training generalist home robots.

Simulation platform with 1,000 everyday household activities for training and testing home robots.

Large dataset of robot arm manipulation across 24 environments, made for learning general skills.

Benchmark and dataset for teaching a simulated robot arm to follow long chains of language instructions.

Large dataset of hand-object contact points plus 2.9 million RGB-D grasp images.

Ready to download dexterous hand manipulation datasets generated by NVIDIA's DexMimicGen pipeline.

Benchmark for capturing how human hands grasp YCB objects, used to train robot grasping models.

NVIDIA's toolkit for loading, visualizing, and benchmarking on the DexYCB hand grasping dataset.

In-the-wild robot manipulation dataset with 76,000 demonstrations across many homes and offices.

Code used to collect the DROID robot dataset, useful if you want to record your own similar data.

Training and evaluation code for learning robot policies directly from the DROID dataset.

Code and tools for loading and using the GRAB whole-body human grasping dataset.

Whole body motion dataset capturing how humans grasp and use everyday objects, for robot grasping research.

4D egocentric dataset of humans interacting with objects, useful for teaching robots grasping skills.

Human collected dataset and benchmark for teaching robots to follow open ended language commands.

Documentation for the LIBERO lifelong robot learning benchmark and its task suites.

Benchmark for testing how well a robot policy keeps old skills while learning new ones.

Hugging Face hosted copy of ManiSkill simulation demonstrations, ready to stream for training.

Guide to downloading and using the large library of ManiSkill robot demonstration datasets.

Fast GPU simulator and benchmark for robot manipulation, with large sets of demonstration data.

Documentation and overview for MimicGen, which multiplies a few demos into thousands of new ones.