The dataset viewer should be available soon. Please retry later.
- Trajectories
- Configs / how to load
- Timestamps
- Minimal example: events around a given frame
- Calibration
- Sensors
- Simulated data (
sim_3d_*,sim_zerog_*configs) - Derived copies of the real recordings (
isolated_*,coregistered,isolated_coregisteredconfigs) - Simulated replay of real recordings (
twin_replay_*configs)
oshd
Synchronized recordings from one rig: a Prophesee EVK4 (IMX636) event
camera, a FLIR Blackfly S Firefly frame camera, an Intel RealSense D455
(color + depth), plus -- where recorded -- a Livox Mid-360 lidar + IMU,
motion-capture poses and the robot's own control/state topics. Everything
is recorded on a single host per trajectory, so every timestamp_ns within
one trajectory shares one clock -- just match by nearest timestamp.
This dataset accumulates multiple recording sessions ("trajectories"), each
with its own full set of sensor streams. Modality is the Hub config; a
split is a group of trajectories (the table below lists each trajectory's
split), and every row has a trajectory_id column. Files are stored one
folder per trajectory, <modality>/<trajectory_id>/. Not every trajectory
necessarily has every modality (a sensor can be off/disconnected for a given
recording); see the table below for exactly which trajectory has which.
Trajectories
| trajectory | split | mockup | light (K) | light intensity | EVK4 temp. (°C) | EVK4 illum. (lux) | events | firefly | realsense_color | realsense_depth | lidar | imu | mocap | robot_messages |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| cubsat_gotopose_com | train | — | — | — | — | — | 141,407,497 | 4,000 | 859 | 1,505 | — | — | — | — |
| cubsatagotopose | train | — | — | — | — | — | 16,048,824 | 1,970 | 366 | 743 | — | — | — | — |
| gopose_test_cubsat | train | — | — | — | — | — | 12,630,096 | 2,192 | 406 | 825 | — | — | — | — |
| gotoposecubsat_final | train | — | — | — | — | — | 37,893,532 | 4,830 | 896 | 1,817 | — | — | — | — |
| init_cubesat_test | train | — | — | — | — | — | 93,883,279 | 9,487 | 3,414 | 3,534 | — | — | — | — |
| test2cubestat | train | — | — | — | — | — | 69,233,531 | 2,762 | 889 | 1,035 | — | — | — | — |
| trajectory_001 | train | — | — | — | — | — | 442,986,965 | 420 | — | — | — | — | — | — |
| trajectory_002 | train | — | — | — | — | — | — | — | 151 | 158 | — | — | — | — |
| trajectory_003 | train | — | — | — | — | — | 1,350,684,260 | 459 | 172 | 178 | — | — | — | — |
| trajectory_004 | train | — | — | — | — | — | 18,333,528 | 1,341 | 296 | 497 | 336 | 627 | 17,971 | 6,686 |
| trajectory_006 | train | — | — | — | — | — | 25,127,922 | 1,635 | 300 | 607 | 409 | 700 | 17,883 | 9,909 |
| trajectory_007 | train | — | — | — | — | — | 24,151,820 | 1,626 | 297 | 593 | 409 | 820 | 26,278 | 9,601 |
| trajectory_008 | train | — | — | — | — | — | 126,755,280 | 1,634 | 302 | 609 | 391 | 918 | 18,897 | 10,417 |
| trajectory_009 | train | — | — | — | — | — | 111,366,953 | 1,637 | 593 | 610 | 367 | 879 | 28,434 | 10,290 |
| trajectory_010 | train | antena | 2750 | 80.2 % | — | — | 486,453,454 | 1,633 | 1,218 | 568 | 4,218 | 517 | 23,015 | 10,617 |
| trajectory_011 | train | antena | 2750 | 80.2 % | — | — | 156,272,366 | 1,587 | 962 | 959 | 4,184 | 783 | 15,455 | 8,518 |
| trajectory_012 | train | antena | 2750 | 80.2 % | — | — | 93,737,710 | 1,674 | 987 | 995 | 4,446 | 826 | 29,621 | 8,719 |
| trajectory_013 | train | antena | 6500 | 4 % | — | — | 67,452,409 | 2,062 | 1,218 | 1,328 | 5,458 | 1,086 | 29,097 | 11,172 |
| trajectory_014 | train | antena | 6500 | 4 % | — | — | 34,068,143 | 1,600 | 998 | 1,032 | 4,200 | 805 | 22,659 | 8,484 |
| trajectory_015 | train | antena | 6500 | 4 % | — | — | 42,689,049 | 1,621 | 1,031 | 1,075 | 4,209 | 936 | 28,676 | 8,607 |
| trajectory_016 | train | antena | 4600 | 22.5 % | — | — | 480,019,395 | 1,634 | 1,212 | 820 | 4,179 | 660 | 26,586 | 10,493 |
| trajectory_017 | train | antena | 5300 | 37.4 % | 28 → 29 | -1 → 0 | 390,671,823 | 1,719 | 1,191 | 596 | 433 | 722 | 17,879 | 9,867 |
| trajectory_018 | train | antena | 5300 | 37.4 % | 29 → 29 | 5 → -1 | 81,342,941 | 1,605 | 1,141 | 405 | 408 | 731 | 23,077 | 9,471 |
| trajectory_019 | train | antena | 5300 | 37.4 % | 30 → 29 | 4 → -1 | 167,188,586 | 1,618 | 1,082 | 245 | 222 | 851 | 23,105 | 9,048 |
| trajectory_020 | train | antena | 2900 | 30.5 % | 30 → 30 | -1 → 5 | 77,744,376 | 1,597 | 1,053 | 1,044 | 237 | 829 | 30,160 | 11,066 |
| trajectory_021 | train | antena | 2900 | 30.5 % | 30 → 30 | 0 → 0 | 368,643,002 | 1,614 | 1,092 | 515 | 408 | 788 | 31,144 | 11,306 |
| trajectory_022 | train | antena | 3550 | 77 % | 30 → 30 | 0 → 0 | 481,431,494 | 1,602 | 1,010 | 717 | 409 | 782 | 23,017 | 10,822 |
| trajectory_023 | train | antena | 5700 | 23.2 % | 30 → 30 | 0 → 0 | 306,035,756 | 1,614 | 1,081 | 770 | 268 | 735 | 24,475 | 11,165 |
| trajectory_024 | train | antena | 5700 | 23.2 % | 30 → 30 | 0 → 0 | 270,454,616 | 1,829 | 1,202 | 959 | 299 | 935 | 19,810 | 10,049 |
| trajectory_025 | train | antena | 5700 | 23.2 % | 30 → 30 | 0 → 0 | 227,690,537 | 1,605 | 1,111 | 1,044 | 409 | 870 | 23,026 | 8,998 |
| trajectory_026 | train | antena | 3050 | 92.6 % | 30 → 30 | 0 → -1 | 464,337,872 | 1,616 | 1,102 | 68 | 409 | 854 | 20,602 | 9,248 |
| trajectory_027 | train | antena | 3050 | 92.6 % | 30 → 30 | 0 → 0 | 259,661,915 | 1,615 | 1,103 | 641 | 302 | 732 | 29,757 | 9,338 |
| trajectory_028 | train | antena | 3050 | 92.6 % | 30 → 30 | 0 → 0 | 371,430,694 | 1,620 | 1,116 | 86 | 317 | 650 | 19,772 | 9,260 |
| trajectory_029 | train | antena | 6200 | 9 % | 30 → 30 | -1 → -1 | 220,011,885 | 1,613 | 1,117 | 585 | 327 | 899 | 24,936 | 9,269 |
| trajectory_030 | train | antena | 6200 | 1.5 % | 30 → 30 | — | 140,167,170 | 1,620 | 1,080 | 1,049 | 409 | 880 | 23,583 | 11,280 |
| trajectory_031 | train | antena | 6200 | 1.5 % | 30 → 30 | — | 209,121,215 | 1,614 | 1,043 | 867 | 409 | 782 | 25,867 | 9,919 |
| trajectory_032 | train | antena | 6200 | 1.5 % | 30 → 30 | — | 91,589,114 | 1,617 | 1,112 | 1,088 | 254 | 834 | 24,040 | 8,263 |
| trajectory_033 | train | antena | 5500 | 6.5 % | 30 → 30 | — | 73,849,269 | 1,621 | 1,141 | 1,162 | 270 | 849 | 25,984 | 9,372 |
| trajectory_034 | train | antena | 5500 | 6.5 % | 30 → 30 | — | 86,192,391 | 1,700 | 1,148 | 1,172 | 389 | 840 | 16,840 | 9,692 |
| trajectory_035 | train | antena | 5500 | 6.5 % | 30 → ? | — | 234,392,931 | 1,620 | 1,109 | 762 | 410 | 784 | 25,202 | 9,592 |
| trajectory_036 | train | antena | 5500 | 6.5 % | 30 → 30 | — | 178,891,723 | 1,635 | 1,142 | 1,044 | 363 | 751 | 24,790 | 9,723 |
| trajectory_037 | train | antena | 3450 | 31.4 % | 30 → 30 | — | 261,336,705 | 1,633 | 1,152 | 776 | 275 | 805 | 31,444 | 9,179 |
| trajectory_038 | train | antena | 4900 | 17.8 % | ? → 30 | — | 136,197,777 | 1,622 | 1,139 | 1,075 | 339 | 750 | 24,007 | 9,243 |
| trajectory_039 | train | antena | 3350 | 51.7 % | — | — | 206,328,881 | 1,619 | 1,107 | 866 | 409 | 771 | 30,505 | 8,892 |
| trajectory_040 | train | antena_1_solar | 5300 | 37.4 % | 24 → 25 | — | 261,022,762 | 1,634 | 1,219 | 399 | 409 | 452 | 20,239 | 10,561 |
| trajectory_041 | train | antena_1_solar | 5300 | 37.4 % | 26 → 26 | — | 233,031,650 | 1,636 | 1,221 | 923 | 409 | 415 | 22,013 | 10,584 |
| trajectory_042 | train | antena_1_solar | 5300 | 37.4 % | 27 → 27 | — | 374,864,071 | 1,628 | 1,219 | 324 | 391 | 566 | 36,350 | 10,495 |
| trajectory_043 | train | antena_1_solar | 5300 | 5.7 % | 27 → 27 | — | 291,203,792 | 1,636 | 1,226 | 400 | 409 | 8,181 | 36,613 | 10,515 |
| trajectory_044 | train | antena_1_solar | 5300 | 5.7 % | 28 → 28 | — | 186,562,057 | 1,636 | 1,222 | 253 | 408 | 8,184 | 26,378 | 10,494 |
| trajectory_045 | train | antena_1_solar | 5300 | 5.7 % | 28 → ? | — | 183,505,870 | 4,826 | 3,461 | 2,079 | 993 | 2,190 | — | — |
| trajectory_046 | train | antena_1_solar | 5400 | 84.3 % | 28 → 29 | — | 175,629,321 | 1,637 | 1,225 | 267 | 410 | 8,187 | 38,763 | 10,568 |
| trajectory_047 | train | antena_1_solar | 5400 | 84.3 % | 29 → 29 | — | 309,968,787 | 1,656 | 1,239 | 610 | 412 | 587 | 29,284 | 10,641 |
| trajectory_048 | train | antena_1_solar | 6050 | 22.9 % | 29 → 29 | — | 106,494,696 | 1,658 | 1,231 | 600 | 414 | 8,293 | 37,818 | 10,668 |
| trajectory_049 | train | antena_1_solar | 4750 | 90.7 % | 29 → 30 | — | 261,981,854 | 1,657 | 1,240 | 1,121 | 414 | 513 | 24,693 | 13,658 |
| trajectory_050 | train | antena_1_solar | 2950 | 95.5 % | 30 → 30 | — | 379,178,987 | 1,634 | 1,218 | 772 | 409 | 540 | 39,001 | 11,441 |
| trajectory_051 | train | antena_1_solar | 4650 | 48.8 % | 30 → 30 | — | 340,086,687 | 1,633 | 1,225 | 795 | 409 | 8,178 | 37,827 | 10,933 |
| trajectory_052 | train | antena_1_solar | 4450 | 18.4 % | 23 → 23 | — | 72,860,362 | 1,637 | 1,203 | 1,165 | 409 | 677 | 38,576 | 10,733 |
| trajectory_053 | train | antena_1_solar | 4450 | 18.4 % | 24 → 24 | — | 252,452,056 | 1,636 | 1,227 | 853 | 409 | 8,181 | 29,202 | 11,058 |
| trajectory_054 | train | antena_1_solar | 4450 | 3.9 % | 24 → 25 | — | 7,311,363 | 1,634 | 1,219 | 1,208 | 409 | 690 | 28,776 | 11,052 |
| trajectory_055 | train | antena_1_solar | 4450 | 3.9 % | 25 → 26 | — | 90,222,326 | 1,630 | 1,216 | 1,063 | 409 | 1,213 | 30,884 | 11,283 |
| trajectory_056 | train | antena_1_solar | 4450 | 3.9 % | 26 → 26 | — | 155,759,413 | 1,635 | 1,219 | 1,100 | 408 | 7,993 | 29,523 | 10,336 |
| trajectory_057 | train | antena_1_solar | 3700 | 23.9 % | 27 → 27 | — | 507,706,948 | 1,634 | 1,223 | 935 | 409 | 8,178 | 28,924 | 10,640 |
| trajectory_058 | train | antena_1_solar | 3700 | 23.9 % | 27 → 27 | — | 168,805,903 | 1,634 | 1,211 | 1,168 | 409 | 3,661 | 38,428 | 10,898 |
| trajectory_059 | train | antena_1_solar | 3700 | 23.9 % | 27 → 27 | — | 87,772,836 | 1,657 | 1,233 | 1,184 | 412 | 8,212 | 38,237 | 10,587 |
| trajectory_060 | train | antena_1_solar | 3700 | 23.9 % | 28 → 28 | — | 25,004,482 | 1,655 | 1,231 | 1,225 | 414 | 8,274 | 28,674 | 10,711 |
| trajectory_061 | train | antena_1_solar | 3700 | 23.9 % | 28 → 28 | — | 346,996,009 | 1,632 | 1,227 | 919 | 402 | 550 | 38,643 | 11,828 |
| trajectory_062 | train | antena_1_solar | 5050 | 67.3 % | 28 → 28 | — | 340,316,127 | 1,636 | 1,226 | 596 | 405 | 483 | 38,703 | 12,946 |
| trajectory_063 | train | antena_1_solar | 5050 | 15.4 % | 28 → 29 | — | 248,363,569 | 1,635 | 1,225 | 762 | 409 | 8,181 | 28,679 | 11,973 |
| trajectory_064 | train | antena_1_solar | 5050 | 15.4 % | 29 → 29 | — | 246,334,847 | 1,637 | 1,225 | 717 | 408 | 491 | 19,170 | 11,059 |
| trajectory_065 | train | antena_1_solar | 5050 | 15.4 % | 29 → 29 | — | 98,005,427 | 1,633 | 1,223 | 1,022 | 410 | 8,187 | 19,157 | 10,749 |
| trajectory_066 | train | antena_1_solar | 5050 | 15.4 % | 29 → 29 | — | 101,384,150 | 1,657 | 1,238 | 1,055 | 415 | 7,039 | 29,367 | 11,211 |
| trajectory_067 | train | antena_1_solar | 6500 | 14.4 % | 29 → 29 | — | 49,695,766 | 1,654 | 1,238 | 1,148 | 414 | 8,275 | 19,470 | 12,581 |
| trajectory_068 | train | antena_1_solar | 6500 | 14.4 % | 29 → 29 | — | 23,114,183 | 1,656 | 1,233 | 1,186 | 415 | 6,208 | 38,414 | 12,306 |
| trajectory_069 | train | antena_1_solar | 6500 | 14.4 % | 29 → 30 | — | 235,055,784 | 1,648 | 1,231 | 805 | 413 | 8,238 | 28,871 | 11,251 |
| trajectory_070 | train | antena_1_solar | 6500 | 14.4 % | 29 → 30 | — | 199,451,068 | 1,642 | 1,225 | 878 | 414 | 5,973 | 34,015 | 10,957 |
| trajectory_071 | train | antena_1_solar | 6500 | 14.4 % | 30 → 30 | — | 187,406,722 | 1,654 | 1,239 | 1,029 | 414 | 3,323 | 38,998 | 10,804 |
| trajectory_072 | train | antena_1_solar | 3000 | 11.7 % | 30 → 30 | — | 135,262,044 | 1,653 | 1,230 | 1,070 | 413 | 6,141 | 28,891 | 11,718 |
| trajectory_073 | train | antena_1_solar | 3000 | 11.7 % | 30 → 30 | — | 410,994,050 | 1,655 | 1,242 | 570 | 414 | 8,079 | 20,263 | 11,207 |
| trajectory_074 | train | antena_1_solar | 3000 | 11.7 % | 30 → 30 | — | 382,311,314 | 1,656 | 1,242 | 675 | 414 | 8,225 | 29,440 | 11,110 |
| trajectory_075 | train | antena_1_solar | 3000 | 11.7 % | 30 → 30 | — | 127,536,864 | 1,634 | 1,222 | 995 | 409 | 8,181 | 19,339 | 10,996 |
| trajectory_076 | train | antena_1_solar | 3000 | 11.7 % | 30 → 30 | — | 69,851,190 | 1,636 | 1,224 | 1,135 | 409 | 8,133 | 19,101 | 12,756 |
| trajectory_077 | train | antena_1_solar | 3000 | 11.7 % | 30 → 30 | — | 87,686,475 | 1,636 | 1,220 | 1,094 | 410 | 8,031 | 24,626 | 12,661 |
| trajectory_078 | train | antena_1_solar | 3350 | 32.2 % | 30 → 30 | — | 94,266,499 | 1,636 | 1,225 | 1,194 | 409 | 501 | 30,014 | 11,665 |
| trajectory_079 | train | antena_1_solar | 3350 | 32.2 % | 30 → 30 | — | 61,510,985 | 1,633 | 1,226 | 1,175 | 409 | 8,182 | 38,873 | 10,601 |
| trajectory_080 | train | antena_1_solar | 3350 | 32.2 % | 30 → 30 | — | 107,526,770 | 1,637 | 1,218 | 1,153 | 408 | 8,094 | 34,080 | 10,554 |
| trajectory_081 | train | antena_1_solar | 3350 | 32.2 % | 30 → 30 | — | 183,933,025 | 1,635 | 1,224 | 847 | 410 | 8,181 | 27,184 | 10,608 |
| trajectory_082 | train | antena_1_solar | 3350 | 32.2 % | 30 → 30 | — | 253,124,883 | 1,634 | 1,217 | 753 | 409 | 8,178 | 28,804 | 12,172 |
| trajectory_083 | train | antena_1_solar | 4000 | 13 % | 30 → 30 | — | 168,547,541 | 1,637 | 1,227 | 866 | 409 | 8,170 | 19,222 | 10,977 |
| trajectory_084 | train | antena_1_solar | 4000 | 13 % | 30 → 30 | — | 427,845,528 | 1,635 | 1,226 | 869 | 409 | 8,183 | 38,976 | 10,908 |
| trajectory_085 | train | antena_1_solar | 4000 | 13 % | 30 → 30 | — | 322,302,976 | 1,638 | 1,221 | 1,005 | 409 | 8,189 | 28,814 | 12,863 |
| trajectory_086 | train | antena_1_solar | 4000 | 13 % | 30 → 30 | — | 228,300,249 | 1,653 | 1,239 | 1,014 | 414 | 8,280 | 38,971 | 10,637 |
| trajectory_087 | train | antena_1_solar | 6100 | 90 % | 30 → 30 | — | 476,085,479 | 1,613 | 1,195 | 539 | 392 | 6,458 | 27,901 | 11,546 |
| trajectory_088 | train | antena_1_solar | 6100 | 90 % | 30 → 30 | — | 188,283,239 | 1,448 | 1,067 | 963 | 354 | 7,112 | 33,525 | 9,133 |
| trajectory_089 | train | antena_1_solar | 6100 | 90 % | 30 → 30 | — | 294,179,666 | 1,638 | 1,219 | 990 | 409 | 8,189 | 38,709 | 10,555 |
| trajectory_090 | train | antena_1_solar | 6100 | 90 % | 30 → 30 | — | 537,186,729 | 1,636 | 1,206 | 696 | 380 | 648 | 37,920 | 10,526 |
| trajectory_091 | train | antena_1_solar | 6100 | 90 % | 30 → 30 | — | 299,505,099 | 1,637 | 1,222 | 981 | 405 | 7,979 | 23,297 | 10,568 |
| trajectory_092 | train | antena_1_solar | 3700 | 30 % | 31 → 30 | — | 270,899,012 | 1,635 | 1,223 | 741 | 409 | 8,178 | 38,573 | 10,528 |
| trajectory_093 | train | antena_1_solar | 3700 | 30 % | 31 → 30 | — | 344,880,947 | 1,803 | 1,339 | 812 | 447 | 9,017 | 41,969 | 11,762 |
| trajectory_094 | train | antena_1_solar | 3700 | 30 % | 30 → 31 | — | 229,508,497 | 1,635 | 1,211 | 1,035 | 402 | 2,394 | 26,718 | 10,341 |
| trajectory_095 | train | antena_1_solar | 3700 | 30 % | 31 → 31 | — | 437,862,014 | 1,631 | 1,219 | 778 | 394 | 563 | 38,543 | 10,535 |
| trajectory_096 | train | antena_1_solar | 3700 | 30 % | 30 → 31 | — | 282,464,239 | 1,634 | 1,210 | 1,034 | 410 | 620 | 36,964 | 10,266 |
| trajectory_097 | train | antena_1_solar | 3700 | 30 % | 31 → 31 | — | 434,166,888 | 1,638 | 1,227 | 700 | 410 | 8,190 | 29,130 | 10,603 |
| trajectory_098 | train | antena_1_solar | 3700 | 30 % | 30 → 31 | — | 398,368,185 | 1,659 | 1,243 | 743 | 416 | 510 | 23,257 | 10,983 |
(— means that sensor recorded no data for that trajectory, or for the light
columns that the light settings weren't recorded; counts are event rows /
image frames. Light is the light source's colour temperature and intensity,
set by hand and entered at recording time -- also in manifest.json's
light; the light's pose is the light body in mocap. Mockup is the
target mockup on the rig; its CAD is models/real/<mockup>/ (metres, origin
at the pivot of its rigid body in mocap), and models/real/mockups.json
maps each trajectory to its mockup -- with static_pose (position in m,
orientation as quaternion xyzw, in the world frame of mocap) when the
mockup stood still and its own mocap rows (body = target) are partial or
missing: then use static_pose. EVK4 temp./illum. are the event sensor's
on-chip temperature and illumination measurement, read off the camera when
the recording started → stopped; also in manifest.json's
event_sensor_monitor.)
Configs / how to load
from datasets import load_dataset
# one modality, every trajectory of a split (filter on trajectory_id):
events = load_dataset("r3m3c3/oshd", "events", split="train")
firefly = load_dataset("r3m3c3/oshd", "firefly", split="train")
rs_color = load_dataset("r3m3c3/oshd", "realsense_color", split="train")
rs_depth = load_dataset("r3m3c3/oshd", "realsense_depth", split="train")
lidar = load_dataset("r3m3c3/oshd", "lidar", split="train")
imu = load_dataset("r3m3c3/oshd", "imu", split="train")
mocap = load_dataset("r3m3c3/oshd", "mocap", split="train")
robot_messages = load_dataset("r3m3c3/oshd", "robot_messages", split="train")
# one trajectory: load its folder (only those files are downloaded)
events = load_dataset("r3m3c3/oshd", data_files="events/cubsat_gotopose_com/*.parquet", split="train")
Each image config yields rows with a decoded PIL image column plus a
timestamp_ns column. events yields a flat table (Arrow/Parquet-backed)
with one row per event.
lidar: one row per scan --scan_index,timestamp_ns,header_stamp_ns,bag_timestamp_ns, and one list column per point field (x,y,z,intensity,tag,line,timestampfor the Mid-360), all the same length within a row.imu: one row per sample -- orientation quaternion, angular velocity (rad/s), linear acceleration (g, the Livox driver's unit, not m/s^2).mocap: one row per pose --body(tracked rigid body name), position (m)x, y, zand quaternionqx, qy, qz, qwin the mocap world frame. Filter bybodyfor one object.robot_messages: every other recorded topic (controller commands, joint states, joystick, arm/disarm, ...) -- one row per message withtopic,msgtype, the time columns anddata, the full message as a JSON string (json.loads(row["data"])). Filter bytopic.
timestamp_ns is always on the recording host's clock. Non-camera streams
additionally carry header_stamp_ns (the publisher's own stamp, null if
the message has none) and bag_timestamp_ns (receipt time at the
recorder); timestamp_ns equals the header stamp when that stream's
header clock agreed with the recorder's to within 1 s, else receipt time.
Timestamps
Every timestamp -- events' t column and every image config's
timestamp_ns column -- is an int64 nanosecond Unix epoch timestamp,
consistent within a trajectory (same recording host, same clock, no
NTP/PTP sync needed) but not comparable across trajectories -- each
recording session starts its own clock reference. To align modalities,
find the nearest timestamp within the same trajectory_id only.
Event timestamps: the sensor's own internal clock is a free-running counter
with no absolute epoch reference of its own. Each trajectory anchors it to
epoch time independently, via an offset confirmed stable across two
consecutive event packets at the start of that trajectory's recording
(guards against a stale/buffered backlog burst on the first packet -- see
bag_to_dataset/convert.py). Events' relative timing to each other is
exact (decoded bit-for-bit from the sensor's raw EVT3 stream); the
absolute anchor is accurate to within one packet's capture latency (a few
milliseconds), the same order of uncertainty the other sensors' own
timestamps carry.
Minimal example: events around a given frame
import numpy as np
from datasets import load_dataset
# one trajectory: load its folders (only those files are downloaded)
events = load_dataset("r3m3c3/oshd", data_files="events/cubsat_gotopose_com/*.parquet", split="train")
frames = load_dataset("r3m3c3/oshd", data_files="firefly/cubsat_gotopose_com/*.parquet", split="train")
t0 = frames[100]["timestamp_ns"]
t_col = np.asarray(events["t"]) # loads the t column as one array
lo, hi = np.searchsorted(t_col, [t0 - 10_000_000, t0 + 10_000_000]) # +/- 10ms
window = events[lo:hi] # dict of x, y, p, t lists for that window
(works the same for any other trajectory -- swap the folder name)
Calibration
See calibration/<trajectory_id>.yaml for that trajectory's per-camera
intrinsics (K, D, R, P) -- kept per-trajectory rather than shared, in case
the rig gets recalibrated between recording sessions. Each entry has a
source field: rig calibration accepted <date> (the checkerboard / monitor
calibration of the rig; these entries also give T_realsense_color_cam, the
camera's pose in the RealSense colour optical frame, p_realsense_color =
T @ p_cam, OpenCV optical frames, RealSense mounted upside down with its raw
image as the reference), ros_camera_info (published live by the driver)
or supplement:... (the driver didn't publish real intrinsics; filled in
from a separate calibration pass instead). Check this before trusting a
camera's numbers.
Sensors
| sensor | model | modality |
|---|---|---|
| events | Prophesee EVK4 (IMX636) | event stream (EVT3) |
| firefly | FLIR Blackfly S BFS-U3-63S4C | RGB frames (debayered) |
| realsense_color | Intel RealSense D455 | RGB frames |
| realsense_depth | Intel RealSense D455 | depth (16-bit, aligned to color) |
| lidar | Livox Mid-360 | 3D point cloud scans |
| imu | Livox Mid-360 built-in IMU | orientation, angular velocity (rad/s), linear acceleration (g) |
| mocap | motion capture (vrpn_mocap) | 6-DoF pose per tracked body |
| robot_messages | robot / controller / teleop topics | one row per ROS message, payload as JSON |
Simulated data (sim_3d_*, sim_zerog_* configs)
Simulated counterpart of the real recordings in this repo, from the
Space Robotics Bench
(Isaac Sim), driven by trained go-to-pose policies. Same layout as the real
data -- one folder per modality, one sub-folder per trajectory -- in two
groups. Each config has one train split with all its trajectories; every
row has a trajectory_id, and sim_manifest.json gives each trajectory's scenario
(e.g. orbital/iss/earth).
| group | trajectories | content |
|---|---|---|
sim_3d_<modality>/ |
4325 | orbital inspection in 3D: a free-flying chaser orbits a static space structure (procedural CubeSats, satellite mockup, Venus Express, Lunar Gateway, ISS) against an Earth or black-sky background |
sim_zerog_<modality>/ |
2288 | Zero-G lab digital twin: the planar Floating Platform inspects or docks to a satellite mockup / CubeSat held by the lab's wall or ceiling arm, or floating |
Configs
| modality | sim_3d | sim_zerog | content |
|---|---|---|---|
| rgb | sim_3d_rgb |
sim_zerog_rgb |
RGB frames: image, timestamp_ns, frame_index |
| depth | sim_3d_depth |
sim_zerog_depth |
depth along the optical axis, 16-bit, units of calibration depth.scale_m (mm, cm for far scenes; 0 = no return) |
| semantic | sim_3d_semantic |
sim_zerog_semantic |
8-bit class IDs: 0 background, 1 unlabelled, 2 target, 3 robot arm, 4 spacecraft |
| events | sim_3d_events |
sim_zerog_events |
t, x, y, p (p: 1 ON, 0 OFF), emulated IMX636 (v2e or openEB: calibration events.emulator), pixel-aligned with the camera |
| lidar | sim_3d_lidar |
sim_zerog_lidar |
one row per scan; list columns x, y, z, distance, channel in the lidar frame |
| imu | sim_3d_imu |
sim_zerog_imu |
robot IMU: orientation_*, angular_velocity_* (rad/s), linear_acceleration_* (m/s^2) |
| mocap | sim_3d_mocap |
sim_zerog_mocap |
ground-truth pose + twist per body: x,y,z, qx,qy,qz,qw, vx,vy,vz, wx,wy,wz (world frame) |
| joint_states | — | sim_zerog_joint_states |
per articulation: position, velocity, effort lists (lab arms, platform) |
| steps | sim_3d_steps |
sim_zerog_steps |
one row per policy step: action, observation, reward, reward_terms, goal errors, task_state |
Per trajectory: calibration/<trajectory>.yaml (pinhole intrinsics K, mount
extrinsics; RGB, depth, semantic and events share one pixel-aligned camera) and
metadata/<trajectory>.yaml (scenario parameters, rates, policy, tracking
summary). sim_manifest.json lists every sim trajectory with its group,
scenario, duration and row counts.
Loading
from datasets import load_dataset
# one modality, every trajectory of the group:
rgb = load_dataset("r3m3c3/oshd", "sim_3d_rgb", split="train")
# every trajectory of one scenario (scenarios: sim_manifest.json):
import json
from huggingface_hub import hf_hub_download
m = json.load(open(hf_hub_download("r3m3c3/oshd", "sim_manifest.json", repo_type="dataset")))
iss = [t for t, e in m["trajectories"].items() if e["scenario"] == "orbital/iss/earth"]
rgb_iss = load_dataset("r3m3c3/oshd", data_files=[f"sim_3d_rgb/{t}/*.parquet" for t in iss], split="train")
# one trajectory (only its files are downloaded):
events = load_dataset("r3m3c3/oshd", data_files="sim_3d_events/trajectory_hera2_000001/*.parquet", split="train")
joints = load_dataset("r3m3c3/oshd", data_files="sim_zerog_joint_states/trajectory_hera2_000007/*.parquet", split="train")
Time and frames
- Timestamps (
timestamp_ns, eventt) are simulation time in ns since the episode reset, not Unix epoch; shared by every modality of a trajectory. - Rates: camera 20 Hz, lidar 10 Hz, mocap/IMU/joints 120 Hz, policy 10 Hz
(exact values in
metadata/<trajectory>.yaml,rates_hz). - Quaternions are
xyzw; world frame z-up at the environment origin.
The full trajectory list (group, scenario, duration) is in SIM_DATASET.md and sim_manifest.json.
Derived copies of the real recordings (isolated_*, coregistered, isolated_coregistered configs)
76 of the real trajectories also come in three derived copies, made from the recordings above by
oshd-isolate. Each folder has the layout of the real data (manifest.json, splits.json,
calibration/, models/, <modality>/<trajectory>/*.parquet), plus isolate/: the corrections,
settings and per-trajectory statistics it was made with.
| folder | configs | content |
|---|---|---|
isolated/ |
isolated_<modality> |
every modality with only the target mockup: the lab (walls, rails, cables, the robot arm holding the mockup, ...) removed, as in the simulated black-space scenarios; adds firefly_semantic, realsense_color_semantic (0 background, 2 target, 255 unknown: no pose) |
coregistered/ |
coregistered |
the original recordings in one reference view, the Firefly's at 0.5 of its resolution: per Firefly frame its image, the RealSense colour and depth, the events of the 25 ms before it and the lidar points within 50 ms of it, with the target's CAD depth, semantic mask, the sources that see each pixel (valid) and its pose T_cam_target |
isolated_coregistered/ |
isolated_coregistered |
the isolated copy in that reference view |
In a co-registered row image and realsense_color are RGB JPEGs; the other channels are PNGs:
depth, realsense_depth and lidar uint16 millimetres (0: none); semantic, valid (bits:
target 1, firefly 2, realsense_color 4, realsense_depth 8, events 16, lidar 32), events_positive
and events_negative (counts, saturating at 255) uint8. T_cam_target is the target model's
pose in the reference camera, row-major 4 x 4 in metres (NaN without one). Per trajectory,
isolate/coregistered/<trajectory>.json holds the reference camera (K, width, height),
describes the channels and counts what was placed.
from oshd import OshDataset # oshd >= 1.2.0
ds = OshDataset.download(variant="isolated", trajectories="trajectory_026", modalities=["firefly", "events"])
from datasets import load_dataset
load_dataset("OSHDataset/oshd", "isolated_events", split="train") # every trajectory (filter on trajectory_id)
load_dataset("OSHDataset/oshd", data_files="coregistered/coregistered/trajectory_026/*.parquet", split="train")
Simulated replay of real recordings (twin_replay_* configs)
twin_replay/ holds the zero-G lab replayed in the Space Robotics Bench (Isaac Sim, RTX) along the
motion-captured trajectories of 4 real recordings (trajectory_024, trajectory_026, trajectory_051, trajectory_097), with the
layout of the real data. Every sensor renders at the real recording's own stamps from the tracked
pose of the sensor stack, with the real calibration (intrinsics, distortion, extrinsics); the
recorded mockup stands at its tracked pose and the lab's lamp at its lens. Modalities: events, firefly, imu, mocap, realsense_color, realsense_depth.
Events come from SRB's srb event-camera model at 250 renders per second, its thresholds tuned to
the real Prophesee sensor in this lab; the IMU is derived from the sensor stack's motion capture.
Calibration, motion capture and object models are copied from the real recordings, and every
stamp of the twin is a stamp of the real recording (frames without a tracked pose are left out).
twin/ holds the rows per modality (counts/) and the recipe that re-renders the twin
(recipe/: render schedules, baked wall textures, camera looks, and the simulator and pipeline
commits in provenance.json).
The walls carry textures baked from the real frames of each trajectory. Known differences from
the real recordings: over whole trajectories the twin fires 1.19, 0.96, 0.96 and 0.78 times the
real events (trajectory_024, _026, _051, _097). The mockup's materials are matte where the
real aluminium, ring and foil glint, so its events differ by model (1.25 and 1.11 times the real
for antena, 0.95 and 0.77 for antena_1_solar); in trajectory_026 a cloth-covered object that
motion capture did not track moves through the view and is not modelled; the darkest pixels fire a
larger ON share than the real ones (32 % against 11 %); there is no lidar.
from oshd import OshDataset # oshd >= 1.2.0
twin = OshDataset.download(variant="twin_replay", trajectories="trajectory_024")
real = OshDataset.download(trajectories="trajectory_024")
from datasets import load_dataset
load_dataset("OSHDataset/oshd", "twin_replay_events", split="train")
- Downloads last month
- 7,616