Dataset Viewer
Auto-converted to Parquet Duplicate
record_id
string
source_object
string
source_generation
string
source_sha256
string
institution
string
collection_date
string
scene_id
string
group
string
split
string
segment_index
int64
raw_start
int64
raw_stop
int64
samples
int64
sample_hz
float64
host_start_ms
int64
time_seconds
list
observed
list
reference
list
request_rows
list
4798fa9e556058b4516fb21967552edff326e7b455ede736b70b708ecdec6541
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
1710756699856916
ad7853f5725e4aeb0043c0047ad5be9d715de11951fc737d616b169765ccfd65
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
2
189
635
314
10
1,699,835,481,241
[ 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2, 2.1, 2.2, 2.3, 2.4, 2.5, 2.6, 2.7, 2.8, 2.9, 3, 3.1, 3.2, 3.3, 3.4, 3.5, 3.6, 3.7, 3.8, 3.9, 4, 4.1, 4.2, 4.3, 4.4, 4.5, 4.6,...
[ [ -0.12369491904973984, 0.1604892909526825, 0.2284516543149948, -2.319979667663574, -0.33358243107795715, 2.655216932296753, 0.7517094016075134 ], [ -0.11658687889575958, 0.16067083179950714, 0.22058336436748505, -2.3202426433563232, -0.3331887423992157, 2.655...
[ [ -0.13824978470802307, 0.14026474952697754, 0.2457232028245926, -2.3206043243408203, -0.30646368861198425, 2.637906074523926, 0.7235071659088135 ], [ -0.02991516888141632, 0.1550280898809433, 0.1351093351840973, -2.321476697921753, -0.3000065088272095, 2.6577...
[ 178, 189, 190, 192, 193, 195, 196, 197, 199, 200, 202, 203, 205, 206, 207, 209, 210, 212, 213, 215, 216, 217, 219, 220, 222, 223, 225, 226, 227, 229, 230, 232, 233, 235, 236, 237, 239, 240, 242, 243, 244, 246, 247, 249, 250, 2...
4798fa9e556058b4516fb21967552edff326e7b455ede736b70b708ecdec6541
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
1710756699856916
ad7853f5725e4aeb0043c0047ad5be9d715de11951fc737d616b169765ccfd65
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
3
643
858
152
10
1,699,835,513,193
[ 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2, 2.1, 2.2, 2.3, 2.4, 2.5, 2.6, 2.7, 2.8, 2.9, 3, 3.1, 3.2, 3.3, 3.4, 3.5, 3.6, 3.7, 3.8, 3.9, 4, 4.1, 4.2, 4.3, 4.4, 4.5, 4.6,...
[ [ 0.2770055830478668, 0.5579943060874939, -0.16451238095760345, -1.4080026149749756, -0.27344974875450134, 2.008087635040283, 0.7774350047111511 ], [ 0.27601203322410583, 0.5586519241333008, -0.16345134377479553, -1.409814476966858, -0.27345749735832214, 2.008...
[ [ 0.2785501778125763, 0.5227960348129272, -0.15880705416202545, -1.4394562244415283, -0.29887324571609497, 2.0062804222106934, 0.7710936665534973 ], [ 0.25512799620628357, 0.5628184080123901, -0.13881617784500122, -1.402495265007019, -0.29209768772125244, 2.00...
[ 634, 643, 644, 646, 647, 649, 650, 651, 653, 654, 656, 657, 658, 660, 661, 663, 664, 666, 667, 668, 670, 671, 673, 674, 675, 677, 678, 680, 681, 683, 684, 685, 687, 688, 690, 691, 692, 694, 695, 697, 698, 699, 701, 702, 704, 7...
648f38de3614a9e4c800a83f13d3e2a68738113063c003f2e80da2ee1c79c912
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:36:18_2023/trajectory.h5
1710756699895125
2dc1460537bce112d20c87dae29cb4ecdf59731b766e490e145f1d9ad199d38d
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
0
8
269
183
10
1,699,835,779,278
[ 0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2, 2.1, 2.2, 2.3, 2.4, 2.5, 2.6, 2.7, 2.8, 2.9, 3, 3.1, 3.2, 3.3, 3.4, 3.5, 3.6, 3.7, 3.8, 3.9, 4, 4.1, 4.2, 4.3, 4.4, 4.5, 4.6,...
[ [ 0.0018039532005786896, -0.8447409868240356, 0.02343807928264141, -2.6872129440307617, 0.13491635024547577, 1.7120394706726074, 0.25938668847084045 ], [ -0.0009141355403698981, -0.8348050117492676, 0.029154187068343163, -2.6875157356262207, 0.1349181979894638, ...
[ [ -0.023341428488492966, -0.8381111025810242, 0.04570814222097397, -2.6846768856048584, 0.1419968456029892, 1.7302354574203491, 0.259063720703125 ], [ -0.021194014698266983, -0.7755685448646545, 0.06091383099555969, -2.6626315116882324, 0.14514917135238647, 1....
[ 6, 8, 9, 11, 12, 14, 15, 16, 18, 19, 21, 22, 24, 25, 26, 28, 29, 31, 32, 34, 35, 36, 38, 39, 41, 42, 43, 45, 46, 48, 49, 51, 52, 53, 55, 56, 58, 59, 60, 62, 63, 65, 66, 68, 69, 70, 72, 73, 75, 76, 78, 79, 80, 8...
648f38de3614a9e4c800a83f13d3e2a68738113063c003f2e80da2ee1c79c912
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:36:18_2023/trajectory.h5
1710756699895125
2dc1460537bce112d20c87dae29cb4ecdf59731b766e490e145f1d9ad199d38d
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
1
277
1,002
510
10
1,699,835,798,196
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.01286581065505743,0.4945080578327179,-0.019858328625559807,-1.450096607208252,0.0255946312099695(...TRUNCATED)
[[0.013922948390245438,0.4957147538661957,-0.020819392055273056,-1.4225056171417236,0.04578065127134(...TRUNCATED)
[268,277,278,280,281,283,284,285,287,288,290,291,293,294,295,297,298,300,301,303,304,305,307,308,310(...TRUNCATED)
4711ab5bf28e27f875ac2ea4140f5f683e77451ab384dcaa9041e1d8c878843b
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:42:22_2023/trajectory.h5
1710756699928022
eec237a7c105e9099bbee4ac5494dc4988cb40d3853c08794398ffc8773ead3a
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
0
19
1,059
731
10
1,699,836,144,024
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.04294556751847267,-0.4935051500797272,-0.1642036885023117,-2.6655895709991455,0.1234597489237785(...TRUNCATED)
[[0.043066415935754776,-0.4677983224391937,-0.1461721658706665,-2.6543729305267334,0.141627550125122(...TRUNCATED)
[17,19,20,22,23,25,26,27,29,30,32,33,35,36,37,39,40,42,43,45,46,47,49,50,52,53,55,56,57,59,60,62,63,(...TRUNCATED)
0d39ff55387813bac85c14fc57a4ea59f50550578917148ee075407078e58b24
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:44:09_2023/trajectory.h5
1710756699953932
1b399acc794f26408bde3a631363d679f24de6242666b93fa8c90d90597be0b8
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
1
128
582
319
10
1,699,836,258,768
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.03891982510685921,0.3372085988521576,-0.04432816803455353,-1.691656470298767,-0.1977925449609756(...TRUNCATED)
[[0.03433172404766083,0.3316916227340698,-0.03847377374768257,-1.6737115383148193,-0.187076926231384(...TRUNCATED)
[117,128,129,131,132,134,135,136,138,139,141,142,144,145,146,148,149,151,152,154,155,156,158,159,161(...TRUNCATED)
0d39ff55387813bac85c14fc57a4ea59f50550578917148ee075407078e58b24
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:44:09_2023/trajectory.h5
1710756699953932
1b399acc794f26408bde3a631363d679f24de6242666b93fa8c90d90597be0b8
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
2
590
797
145
10
1,699,836,291,283
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.12680421769618988,0.2975260019302368,0.07548406720161438,-1.5784274339675903,-0.2248422652482986(...TRUNCATED)
[[0.12790866196155548,0.2939703166484833,0.07737871259450912,-1.559818983078003,-0.21225816011428833(...TRUNCATED)
[581,590,591,593,594,596,597,598,600,601,603,604,606,607,608,610,611,613,614,616,617,618,620,621,623(...TRUNCATED)
ae82f4bf8d59f6d039738af2e255207934eeaa4bf72f9f229fae64c18a7253fe
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:45:17_2023/trajectory.h5
1710756699983123
7eedff9e9513b6b63b3b146a90f8f864a0e55a23cc075d74912924cfa24d683a
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
0
7
722
502
10
1,699,836,318,383
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.16114112734794617,-0.5082380771636963,0.22137781977653503,-2.5748848915100098,0.1414773911237716(...TRUNCATED)
[[0.16779720783233643,-0.5079911947250366,0.21515090763568878,-2.5721278190612793,0.1374870836734771(...TRUNCATED)
[5,7,8,10,11,13,14,15,17,18,20,21,23,24,25,27,28,30,31,32,34,35,37,38,40,41,42,44,45,47,48,50,51,52,(...TRUNCATED)
16884fb668c8194684e4bc6a72ecae26c7d0301841f293e9be847368a10fad71
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:46:22_2023/trajectory.h5
1710756700026563
9525822678ec8d1e673200c22ec5efb75ecbd7875022cce4d277e5f4760cce1d
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
2
142
700
392
10
1,699,836,392,419
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.04865372180938721,0.20622193813323975,-0.07597509771585464,-2.0819501876831055,-0.05012599378824(...TRUNCATED)
[[0.043753717094659805,0.1719711422920227,-0.07135757803916931,-2.1101279258728027,-0.03100923262536(...TRUNCATED)
[134,142,143,145,146,148,149,151,152,153,155,156,158,159,160,162,163,165,166,168,169,170,172,173,175(...TRUNCATED)
50d907a1bf0f0339d874b2aef9b2079877dfc77d7f7e792d7187604ae6427cf4
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:51:13_2023/trajectory.h5
1710756700105196
f06ad2e09976770a64c27ed21187e8df887b3ff121caee9642770dbc068ac32c
AUTOLab
2023-11-12
9487913557
robotics/droid_raw/1.0.1/AUTOLab/success/2023-11-12/Sun_Nov_12_16:31:07_2023/trajectory.h5
confirmation_session
0
11
558
384
10
1,699,836,674,422
[0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0,1.1,1.2,1.3,1.4,1.5,1.6,1.7,1.8,1.9,2.0,2.1,2.2,2.3,2.4(...TRUNCATED)
[[0.2775653600692749,-0.7197279334068298,0.03279012814164162,-2.6749744415283203,0.31277593970298767(...TRUNCATED)
[[0.2578281760215759,-0.7024047374725342,0.04709774628281593,-2.668970823287964,0.3309604227542877,1(...TRUNCATED)
[9,11,12,14,15,17,18,19,21,22,24,25,27,28,29,31,32,34,35,37,38,39,41,42,44,45,47,48,49,51,52,54,55,5(...TRUNCATED)
End of preview. Expand in Data Studio

DROID motion for mixture-consistent factorization

Motion-only derivative of DROID, raw release 1.0.1. Prepared by Sam Liu for experiments on separation of robot motion and tracking-error dynamics. This is an independently prepared subset, not an official DROID benchmark or evidence that a robot denoiser works.

The DROID paper, Section III, releases the full dataset under CC BY 4.0. This derivative retains that license. Please credit Alexander Khazatsky, Karl Pertsch, Suraj Nair and the DROID Dataset Team, cite the paper below, link the source dataset and this conversion, and identify any further modifications. No endorsement by the DROID authors is implied.

What changed

Selected low-dimensional HDF5 trajectories were converted to compressed Parquet; images, video, language annotations and collector identifiers are not included. Seven joint positions and requested positions are retained. Native values use float64. The default configuration resamples measured positions to 10 Hz with offline linear interpolation, stores positions as float32, and holds the most recent eligible logged request. It includes every continuous span of at least 128 grid samples (12.7 seconds between first and last samples), not only the longest span. No global normalization is fitted. Units are radians.

Configurations and loading

from datasets import load_dataset

segments = load_dataset(
    "sammyliu/droid-motion-mixit-v1",
    "motion10hz",
    split="train",
    streaming=True,
)
segment = next(iter(segments))
# segment["observed"] and segment["reference"] each have shape [time, 7].

Pin revision= to a verified commit for experiments. motion10hz is the default configuration: one continuous segment per row. native has one validated source recording per row, with measured/requested float64 arrays, host event times in integer milliseconds, skipped-action flags, and hardware-clock seconds/nanoseconds when available (empty arrays otherwise). Native records can contain gaps and skipped steps; these must be handled before training. Native records flagged by duplicate_of are retained for provenance and excluded from motion10hz.

Each row includes the source object, pinned GCS generation, original file SHA-256, institution, date, scene ID, connected group and split. Segment rows additionally include the raw half-open span, relative grid timestamps, host start time and source request indices. provenance/segment-index.parquet maps all segments to shards and row offsets, supporting global random sampling independently of cache state. Sample globally over recordings before selecting segments/windows if equal recording weighting is desired; sampling rows alone weights fragmented records more heavily. Multiple windows from one recording are not independent recordings.

Timing and target meaning

Host events must follow step-start <= read-start <= read-end <= control-start <= step-end. Reads strictly increase and return times are nondecreasing. The reference is the last non-skipped request whose logged control call returned strictly before a read or grid time. Ties do not make a request available. Skipped rows, missing references and read gaps greater than 150 ms split spans. Fragments are never joined, and interpolation never crosses such a boundary.

The reference is a logged requested position, not a verified actuator-applied command, simulator ground truth or independently measured ideal motion. Request return time is not actuator application time. Hardware timestamps use a separate clock and are retained without aligning them by an optimized lag. The 10 Hz representation is offline and can remove high-frequency motion; it is not a causal online sensor stream or an anti-alias-filtered representation. Use native data for alternative rate and filtering studies.

Splits and exclusions

The candidate pool contains successful recordings from nine institutions. RAD, CLVR, IRIS and TRI were excluded as whole held-out institutions. Previously held-out files and identical whole-file checksums were excluded from acquisition. The expanded grouping additionally includes all prior split anchors, even when their recordings are absent here. Institution/day, institution/scene and exact measured/requested float64 hashes define transitive connected groups. Groups inherit a unique prior split; groups connecting conflicting prior splits are placed entirely in quarantine. New unanchored groups receive deterministic 80/10/10 train/development/confirmation_session assignments. These are custom research splits, not DROID's original published splits.

Never include quarantine in a training/evaluation comparison. Keep every group in a single split and preserve these rules when expanding the dataset. Exact motion duplicates are retained only once in motion10hz. Counts describe files, connected groups, segments and non-overlapping windows separately. Nearby windows and even different recordings can be correlated. The whole-institution holdout payload is not included in this release.

Measured population

The first three counts below are recordings: candidates, native validated, and distinct recordings admitted to the 10 Hz configuration.

Split Candidates Native Admitted Groups Segments Hours 128-sample windows
confirmation_session 2377 2376 1071 23 1408 9.232 2092
development 3573 3572 1457 23 1551 9.566 2160
quarantine 1770 1770 1176 5 1573 10.182 2304
train 23401 23396 8341 194 10604 79.274 18384

provenance/summary.json contains context-length counts through 2,048 samples, processing parameters and source/split fingerprints. split-contract.json includes every selected source and exclusion reason. SHA256SUMS.json covers every published data/provenance file and this card. The raw HDF5 staging files are not uploaded; source object generations and hashes permit re-acquisition.

Citation

@inproceedings{khazatsky2024droid,
  title = {DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset},
  author = {Khazatsky, Alexander and Pertsch, Karl and Nair, Suraj and others},
  booktitle = {Robotics: Science and Systems},
  year = {2024},
  url = {https://droid-dataset.github.io/}
}
Downloads last month
108