datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
DreamZero-DROID-Datadreamzero-egoverse-360-pretraindreamzero-egoverse-rawdreamzero-franka-pnp
DreamZero Franka Pick-and-Place
This dataset contains real-world Franka Panda pick-and-place demonstrations used for DreamZero supervised fine-tuning and real-world deployment in RLinf.
Dataset Summary
Format: LeRobot v2.1
Robot: Franka Panda
Task: Pick up the object and put it into another bin
Episodes: 140
Frames: 9,685
FPS: 10
Videos: 280
Image streams: observation.images.image, observation.images.wrist_image
Image size: 128 x 128 RGB
State: float32, shape… See the full description on the dataset page: https://huggingface.co/datasets/RLinf/dreamzero-franka-pnp.bimanual-ur3-dreamzero
Bimanual UR3 Dataset (DreamZero / LeRobot v2)
Dual UR3 arms with Robotiq 2F-85 grippers, three RGB cameras (main + left/right wrist).
Prepared for DreamZero embodiment fine-tuning.
Stats
Episodes: 106
FPS: 30.0
State/action dim: 14 (6+1 per arm)
Embodiment tag: bimanual_ur3
Layout
data/chunk-000/episode_XXXXXX.parquet
videos/chunk-000/observation.images.*/episode_XXXXXX.mp4
meta/modality.json, stats.json, ...
Training… See the full description on the dataset page: https://huggingface.co/datasets/JeffrinSam/bimanual-ur3-dreamzero.bimanual-ur3-dreamzero-v2
Bimanual UR3 Dataset for DreamZero (v2)
LeRobot v2.1 + GEAR metadata for DreamZero embodiment fine-tuning.
v2 changelog (use this version)
Fixed actions: action = observation.state (original logs had zero actions)
Task prompt: pick_up_greenball_using_rightarm_and_pass_it_to_rightarm
106 episodes, 43,417 frames, 30 FPS
3 cameras: main (1280x720), left_wrist, right_wrist (640x480)
14-D bimanual state/action (6 joint + 1 gripper per arm)
Layout… See the full description on the dataset page: https://huggingface.co/datasets/JeffrinSam/bimanual-ur3-dreamzero-v2.dreamzero-egoverse-480-pretrainDreamZero-DROID-Dataroborag-dreamzero-fixed-video-warmup
RoboRAG fixed video warm-up references for DreamZero
This dataset contains the three fixed DROID reference trajectories used in
the retained DreamZero fixed-reference experiments, plus a small server-side
Method-V injection adapter. It does not include a DreamZero checkpoint,
robot policy weights, simulator, or robot driver. Load the DreamZero
checkpoint you are authorized to use in your own checkout.
The three references are synchronized 4-s, 15-FPS, three-view DROID clips:… See the full description on the dataset page: https://huggingface.co/datasets/YunzeLiu/roborag-dreamzero-fixed-video-warmup.libero-spatial-dreamzero
LIBERO-Spatial Dataset for DreamZero
This dataset contains 500 expert demonstrations from the LIBERO-Spatial benchmark, preprocessed and ready to use directly with DreamZero for training.
Dataset Overview
Property
Value
Format
LeRobot v2.0
Robot
Franka Panda
Total Episodes
500
Total Frames
62,250
FPS
20
Image Resolution
128×128
Camera Views
agentview, eye_in_hand
Tasks
10 (LIBERO-Spatial suite)
Tasks (10 Spatial Arrangement Tasks)… See the full description on the dataset page: https://huggingface.co/datasets/huiliu123/libero-spatial-dreamzero.DreamZero-DROID-Data-PicknPlaceDreamZero-DROID-Data-2000vllm-omni-dreamzero-assets
vLLM-Omni DreamZero Example Assets
Small camera videos used by the vLLM-Omni DreamZero OpenPI example.
Files:
exterior_image_1_left.mp4
exterior_image_2_left.mp4
wrist_image_left.mp4
dreamzero_resultdreamzero-egodex-rawdreamzerodreamzero-egodex-pretraindreamzero-egoverse-360-pretraindreamzero-egoverse-480-pretraindreamzero-egodex-pretrain