ALOHADownloadALOHA Bimanual Teleoperation Dataset
Download ALOHA bimanual teleoperation datasets for robot learning. Sim and real-world manipulation demos in LeRobot format. Apache 2.0 licensed. Free to use.
Search, preview, and download robotics data — manipulation, locomotion, tactile sensing, motion capture, and more. License-aware access: we respect every dataset's terms.
Datasets with in-the-wild robot interactions and long-horizon tasks.
→CollectionSuites designed for reproducible evaluation and cross-paper comparison.
→CollectionShared formats and multi-embodiment data for foundation model training.
→Built for users who search by workflow, industry, or decision intent rather than by a single named dataset.
Operator demos, retries, and bootstrapping workflows.
→Dataset GuideTactile, force, and failure-heavy manipulation signals.
→Industry GuideSKU variation, exception handling, and throughput context.
→Industry GuideRepeatable protocols and benchmarkable workflows.
→Pilot GuideDeployment-oriented data choices for humanoid teams.
→OpenArm GuideCollection and packaging workflows around OpenArm.
→47 datasets match current filters.
ALOHADownloadDownload ALOHA bimanual teleoperation datasets for robot learning. Sim and real-world manipulation demos in LeRobot format. Apache 2.0 licensed. Free to use.
Stanford / NVIDIADownloadALOHA bimanual teleoperation datasets: 50+ demos per fine-grained task, open hardware reference. Train ACT/Diffusion Policy in 30 minutes — download now.
AMASSLink OnlyAMASS: the largest unified human motion capture dataset. 40+ hours of motion from 15 optical marker-based datasets, fitted to the SMPL body model. Academic license.
Assembly101Link OnlyAssembly101: 513 hours of assembly/disassembly video from 53 participants, 12 synchronized cameras, 1M+ fine-grained action annotations. CC-BY-NC-4.0.
2026 GuideLink OnlyThe 10 best robot learning datasets for 2026 ranked. Open X-Embodiment, DROID, BridgeData, LIBERO, CALVIN, ALOHA and more, with scale, license, and use cases.
CollectionLink OnlyCurated bimanual manipulation datasets for robot learning: ALOHA, Mobile ALOHA, DROID bimanual splits, and custom dual-arm collections. Download or commission new data from RCSV.
2023DownloadBridgeData V2: 60K trajectories, 24 environments, 13 skills. Low-cost WidowX manipulation dataset from Stanford.
2023DownloadDownload BridgeData V2: 60,096 WidowX trajectories across 24 environments and 13 skills. Train language-conditioned policies and benchmark generalization in under 1 day.
ComparisonLink OnlyBridgeData V2 vs RoboMimic: real vs sim, trajectory counts, task suites, language conditioning, and hardware compared. Pick the right manipulation dataset for 2026.
CALVINLink OnlyExplore CALVIN: 24 hours of teleop across 34 tasks in 4 environments with natural language labels. Download, fine-tune HULC, and benchmark in 15 minutes.
CMUDownloadCMU Motion Capture Database: 2,605 motion sequences across 6 categories. Free for all uses. The most widely used motion capture dataset in robotics and animation.
Dataset GuideLink OnlyDatasets for contact-rich manipulation, insertion, force-sensitive control, tactile reasoning, and failure-heavy robot learning tasks.
RCSVLink OnlyOrder bespoke imitation learning data: pick the robot, tasks, modalities, and scale. 24-hour teleoperation coverage at our San Francisco lab. From $500/task episode. Get a quote in 24 hours.
RSS 2024Link OnlyDROID: 76K trajectories, 350 hours, 86 tasks across 564 real scenes on Franka Panda. Download via TFDS/HuggingFace and fine-tune OpenVLA in an afternoon.
Meta AILink OnlyEgo4D: the largest egocentric video dataset. 3,670 hours from 931 participants across 9 countries. Hand tracking, gaze, 3D scans. Research use only (CC-BY-NC-4.0).
EPIC-KITCHENSLink OnlyEPIC-KITCHENS-100: 100 hours of unscripted egocentric kitchen activities from 45 kitchens. 90K action segments. Standard benchmark for action recognition and anticipation.
Dataset GuideLink OnlyEvaluation datasets for robotics teams that need benchmarkable scenarios, repeatable resets, and deployment-oriented testing loops.
Dataset GuideLink OnlyWhy failure replay datasets matter for robotics, and how retries, interventions, and outcome labels improve learning and evaluation.
Dataset GuideLink OnlyHow to evaluate datasets for robotics foundation model training, including multi-robot coverage, language conditioning, and action standardization.
MIT CSAILLink OnlyGelSight tactile sensor datasets from MIT CSAIL. High-resolution vision-based touch data for grasp planning, texture recognition, and slip detection. Research license.
Meta AILink OnlyHabitat simulation platform data: photorealistic 3D indoor environments for embodied AI navigation, rearrangement, and mobile manipulation. Meta AI. Mixed licenses.
HumanML3DDownloadHumanML3D: 14,616 human motions with 44,970 text descriptions. The standard benchmark for text-to-motion generation. MIT licensed, built on AMASS + HumanAct12.
CollectionLink OnlyCurated humanoid robot datasets for locomotion, whole-body manipulation, and loco-manipulation policy training. Open-source and custom collection from RCSV.
Dataset GuideLink OnlyHumanoid robotics datasets for mobility, manipulation, operator interventions, benchmark gating, and deployment-oriented learning workflows.
Industry GuideLink OnlyLab automation datasets for sample handling, vial transfer, repeatable resets, and benchmarkable robotics workflows in life sciences settings.
Hugging FaceDownloadBrowse LeRobot open-source robotics datasets: DROID-100, ALOHA bimanual, SO-100. Standardized Parquet + MP4 format, hosted on Hugging Face Hub with PyTorch streaming support.
Hugging FaceDownloadBrowse the LeRobot dataset hub by HuggingFace: 181+ standardized robotics datasets in Parquet+MP4 format. ALOHA, DROID, PushT, xArm, and more. Apache 2.0 framework.
ComparisonLink OnlyLIBERO vs CALVIN in 2026: task suites, demos, language conditioning, embodiments, and licenses compared. Pick the right benchmark for lifelong or long-horizon robot learning.
BenchmarkDownloadExplore LIBERO: 130 tasks, 65K demos across 4 lifelong learning suites. Download, train OpenVLA/Diffusion Policy, and benchmark transfer. Get started in 10 minutes.
CollectionLink OnlyCurated robot manipulation datasets for policy learning: DROID, BridgeData v2, Open X-Embodiment, RoboMimic, and custom single-arm collections. Browse, compare, and download.
ManiSkill2DownloadManiSkill2: GPU-accelerated manipulation benchmark with 20+ task environments, expert demonstrations in HDF5 format. Apache 2.0 licensed. RGB, depth, point cloud modalities.
NVIDIADownloadExplore MimicGen: turn 200 human demos into 50,000+ synthetic demonstrations. Install NVLabs MimicGen, scale imitation learning 250x, and benchmark Diffusion Policy today.
Dataset GuideLink OnlyA guide to multimodal robot datasets with RGB, depth, force, tactile, proprioception, and language annotations for embodied AI.
NVIDIALink OnlyNVIDIA GR00T teleoperation datasets for humanoid robots. GR1, G1, sim variants. 7.55M+ rows. Foundation model training data for physical AI. NVIDIA Open Model License.
ComparisonLink OnlyOpen X-Embodiment vs DROID: trajectory scale, embodiments, data format, license, and training compute compared. Pick the right foundation-model robot dataset in 2026.
Google DeepMindDownloadExplore Open X-Embodiment: 1M+ trajectories across 22 robot embodiments and 527 skills. Download RT-X data, fine-tune VLAs, and benchmark cross-embodiment transfer in hours.
Reazon ResearchDownloadOpenArm dataset and data workflow guide covering teleoperation, collection packaging, failure traces, and training-ready delivery.
Google ResearchDownloadReal-World RL Suite from Google: standardized benchmarks for reinforcement learning on physical robots. Apache 2.0 licensed. Sim-to-real transfer and safety constraints.
Dataset GuideLink OnlyCompare real-world and simulation datasets for robotics, including where each is strong and where teams usually need blended data strategies.
RH20TLink OnlyRH20T (Real Human 20 Tasks): 110K+ contact-rich manipulation sequences across 7 robot configs with tactile, force-torque, audio, and multi-camera data. 5TB+.
RoboCasaDownloadRoboCasa: large-scale kitchen manipulation dataset with 536K+ downloads. MIT licensed simulation data for household robotics. Pretrain generalist kitchen policies.
ARISE InitiativeDownloadExplore Robomimic: 6 task suites, 1000+ demos per task, built on robosuite and MuJoCo. Download, reproduce BC-RNN and Diffusion Policy baselines, and benchmark in 1 afternoon.
Stanford / BerkeleyDownloadExplore RoboNet: 15M video frames across 7 robots and 4 labs. Download, train video prediction and visual foresight models, and benchmark cross-robot transfer in 1 weekend.
Dataset GuideLink OnlyA practical guide to teleoperation datasets for robotics, including what to look for in demonstrations, failure traces, and learning-ready supervision.
Touch and GoLink OnlyTouch and Go: paired visual and tactile data for learning cross-modal representations. Research-only license. Useful for tactile-augmented robot manipulation.
Unitree G1Link OnlyUnitree G1 humanoid robot datasets: 124+ task variants including dishwasher loading, towel folding, toast making, block stacking. Whole-body teleoperation data in LeRobot format.
Industry GuideLink OnlyWarehouse robotics datasets for picking, tote transfer, exception handling, SKU variation, and logistics evaluation workflows.
We highlight scale, format, and access details needed for quick evaluation.
Datasets are mapped to practical model and tool ecosystems.
Dataset choices are linked with real robot execution constraints.
When open data is not enough, we support custom collection pipelines.
We collect high-quality, learning-ready data for your specific tasks and hardware.