running
161,984 / 400,000 episodes saved
Right Sharpa hand · Work in progress
System0
A promptable policy for dexterous manipulation.
We are building a policy for the right Sharpa hand that follows object-motion and hand-pose prompts.
How this is measured
Average completion of the 15 jobs below, weighted equally. Collection uses episode targets; checks, downloads, and training use their own totals. This measures the current work plan, not the hand’s task success.
The pipeline
From demonstrations to control- 01
Collect
Roll out trained policies and gather public hand–object recordings.
- 02
Check
Test right-hand execution. Reject drops and failed tasks; align poses, actions, and contacts.
- 03
Train
Learn a diffusion policy from successful motions and their object-motion references.
- 04
Prompt & control
Provide a motion goal. Predict wrist and finger targets, execute a short segment, then replan.
The current model dataset covers pickup and hold. Broader skills and human-pose retargeting are still being developed.
The data
Choose a dataset to watch an exampleCube pickup and hold1× speed
System0 pickup
Pick up a cube or ball, then hold it above the table.
Floating right hand · The current training set uses these motions.
Original source ↗Full source & checkpoint list 66 entries
66 entries
SimToolRealMirrored tool motion; mallet downward swingCheckpoint
Revision 313d5aea1f507c6cfe097b672b62945d7b0bbff5
Recorded teacher · SHA-256 4cb0d09833326de6878ee5681c2b14521823fa616e01b135214a51ffe41c63b9
DEX-XCube grasp; squeegee manipulationCheckpoint
Revision 88d59f629eb7d74728374523238a78ebded6e943
teacher_poseobs.pth · SHA-256 42c774921e27b7291430ce59c71c8e1c7bf3efad752aa7e50bf42a8ab07d5edd
Sharpa RL LabIn-hand rotationCheckpoint
Revision 95ccda3d948801bb5da4cb7ffea766e03067a63b
Recorded teacher · SHA-256 316dbca720dfb2ec1e0f593b192126a4531d30f3211d236f7a407ec9db15af51
Recorded teacher · SHA-256 c37dff6fd9b204f7bb69cae5eb609180329beee4fdb3573b262228a8e0a6d757
WM-CraftnetIn-hand rotation across nine object categoriesCheckpoint
Revision 7a200ba759f981b7540acffe930d4036bbb3159e
wm_craftnet_set_z.pth · SHA-256 3215205fe1fad48e97f5745ab90bb2377d53012c1a0d2634edf186699ccab7f7
Pen manipulationThree-turn pen spin, stop and holdCheckpoint
Revision 5e60ae21cd29866aefe3e2cffee9b260c2b0a7b1
Recorded teacher · SHA-256 cc88085343402380b07df560a874b7784e05ed9fb8f75e0b19c575c7e08bf4cc
CHORD / Video-to-Data teacherBimanual tissue-box lift, hold and returnCheckpoint
Revision 7eb9721088615fe9201634a6be5cf48b049d1090
vega_sharpa_policy.onnx · SHA-256 bf3259ec2b86dd696e6b89e6275d85985ad30dafeaf53351971b16cdb8dad8b2
Sharpa residual PPOObject manipulation candidateCheckpoint
Revision 63eb446a28d32e2c7c27de85b224396e69475abc
Dexterity from Supervised PlayGoal-directed in-hand cube reorientationCheckpoint
Revision e41cf7c4f394c1e9ca24761b980d1193a5db75ef
dsp_sim_play_policy.pt · SHA-256 744a4159f7842e6e2c7dbac57748c74c771f5c13ffdfa7b302757b6f75b8b463
dsp_sim_world_model.pt · SHA-256 3d337874d91b06f12fac94ab62f51a02772ce819191b2002ecb8dc1e713b1498
SPIDERGrasp, lift and supported hold across diverse objectsTrajectories
Revision 1bb88b0c32b33eb868def58333a516ad2f2aba30
SharpaDex59 published task categoriesTrajectories
Revision 8e04aab1255da8eb73fe2b31516054dbb9e15eda
NVIDIA Video-to-Data1,215 published motion sequencesTrajectories
Revision 2376c2eb477685ac5da47d3b634fbed31c0ccef5
T-Rex demonstrations22 motor primitives; 207 objects; 5,464 published episodesTrajectories
Revision 8b478400a9167d59ede0fdbfea616cb5e9791885
T-Rex midtrained policyTactile-reactive policy initializationCheckpoint
Revision 62efb3bcb45a3df0e088c8909d759b582cfb98af
T-Rex pretrained policyVision-language-action initializationCheckpoint
Revision d9cbd9afd298767137fdb01cc9f96fad0cc6ddb2
DeCAL demonstrations13 tasks; 1,279 published episodesTrajectories
Revision bcf54d0d6c48d0fbaa03fa0d2f27890d37649872
FTP-1 Sharpa subsetTactile manipulation; coverage audit pendingTrajectories
Revision d287c9c532514a08b2eb29bf1259997bb32f4f3b
EigenDEXploreTraining / adaptation toolsCode
Revision 5d57180eb247a8e2e6b809fa019495374317d47d
DeCAL policy codeTraining / adaptation toolsCode
Revision 09de1855cfc933a2d2680d6011744476647ed7d1
X-ResetGoal-conditioned manipulation and recoveryResearch
System0 pickup teachersCube and sphere pickup with supported retentionCheckpoint
Recorded teacher · SHA-256 5f1e316cdf655b722cc7f176c6196d55a1450dc781258a3bd069be297929ca76
Recorded teacher · SHA-256 fbbee39e6cf120419488b6b545cd5e97328a3c4f2d6217bb5a7a9c1296f78948
Bench2Dex/teleopdata · bench2dex-sharpa-tasksTray arrangement, baking preparation and puzzle assembly; source variantsTrajectories
Revision 7c4c2e1cf67683857dbb0b3610cf1d20b1020c55
Bench2Dex/Assets · bench2dex-task-assetsTray arrangement, baking preparation and puzzle assembly; source variantsCode
Revision bf65215f844d1fca30750e4cc70d7266650abf0a
Bench2Dex/AssetsTray arrangement, baking preparation and puzzle assembly; source variantsCode
Revision bf65215f844d1fca30750e4cc70d7266650abf0a
Bench2Dex/teleopdataTray arrangement, baking preparation and puzzle assembly; source variantsTrajectories
Revision 7c4c2e1cf67683857dbb0b3610cf1d20b1020c55
Dex-GEM/DexCanvasHuman hand–object motion, grasp transitions and contactsTrajectories
Revision 69380022d93cb3fb31c9e22c051772a3c5d55af1
FlappyHimself/sharpa-bc-data018-phone18-c4096Phone manipulationTrajectories
Revision 35df07dfc170d5628b172feec2a81d0cec4f3fdf
FlappyHimself/sharpa-bc-data019-phone07-partial2763-powPhone manipulationTrajectories
Revision fb5360dad48390d1ae2fbd938d24d4458443060a
FlexPhysics/sharpa-cylinder-rotationRight-hand cylinder rotation across eight scalesTrajectories
Revision c62815d1619125686b4c2f466fa15bb58f6af0f9
JensenYuan/DexTacWAM_488_diverse_episodesDexTacWAM 488 diverse episodesTrajectories
Revision c9142c56c9c15a7dbb1de9fb394d4af37a5cea94
JensenYuan/DexTacWAM_cube_handoverDexTacWAM cube handoverTrajectories
Revision 00b652dd60b1f718e029ac9d235be60594c1a05b
JensenYuan/DexTacWAM_pick_cherry_tomato_with_tong_rawDexTacWAM pick cherry tomato with tong rawTrajectories
Revision d570f6e8f1d63a7e409f266a8f274137f7f0d9c2
JensenYuan/DexTacWAM_pick_cube_in_a_bag_rawDexTacWAM pick cube in a bag rawTrajectories
Revision 2d0de63cdc82f6d8f0b90c5e1b1aa5109eb0a076
JensenYuan/DexTacWAM_pick_cube_right_hand_rawDexTacWAM pick cube right hand rawTrajectories
Revision b1adbb6394779aa7c4ee6b66ca89aeb7953ac904
JensenYuan/DexTacWAM_pick_place_cubeDexTacWAM pick place cubeTrajectories
Revision f6a08ab77e3b91494cee1cc872d04cc3dc1a2eb3
JensenYuan/DexTacWAM_place_with_tongsDexTacWAM place with tongsTrajectories
Revision 9c79ec40d649b11ef2dce6517f332fa26661e5c7
JensenYuan/DexTacWAM_unscrew_bottle_capDexTacWAM unscrew bottle capTrajectories
Revision 2fb714b635cb13e00b01a3b4c255abd50d510fbe
JensenYuan/DexTacWAM_unstack_bowlDexTacWAM unstack bowlTrajectories
Revision cfbd1074ed5fec6c409987ea364973d29ce6e2f2
JensenYuan/DexTacWAM_wipe_white_board_rawDexTacWAM wipe white board rawTrajectories
Revision 230c84b2e709f760c8f302dbbfc5af21d9712a5a
JensenYuan/DexTacWAM_wipe_whiteboardDexTacWAM wipe whiteboardTrajectories
Revision c688eb4d95e96f4f3587f9a8a52940251da76b0a
Myungkyu/bench2dex_sharpa_lerobot_v21_rldxTray arrangement, baking preparation and puzzle assembly; source variantsTrajectories
Revision 93a35185040cd9bdf328b29371e66c446bc0db1e
Myungkyu/bench2dex_sharpa_lerobot_v30Tray arrangement, baking preparation and puzzle assembly; source variantsTrajectories
Revision 986535ccce64c41dcc76973c6fc8f31f9c31ae13
SharpaIT/RoCo_TaskBoardAssemblyTaskboard assemblyTrajectories
Revision 07432c3ecb3a97a067ca653dcfa751b15f6d8181
Voxel51/sharpa-assemble-gears-410epsharpa-assemble-gears-410epTrajectories
Revision 9f2a17532646dff720a9d4330152c8a743e99873
jeremynut/pico-manus-sharpa-fr3-teleoppico-manus-sharpa-fr3-teleopTrajectories
Revision 2db9b362d1b33ce0880ca60cd9447cb52c60a2f3
zju/sharpa_grasp_bottle22249179_geo_visual100_2camsharpa grasp bottle22249179 geo visual100 2camTrajectories
Revision cc6340e3b2dc7c28b74ef78d09056563466538fd
Bench2Dex/policy_ckptTray arrangement, baking preparation and puzzle assembly; source variantsCheckpoint
Revision 773e65842bd36de1cbe951b0f52c865a87d41732
JensenYuan/DexTacWAM_multi_finger_tactile_encoderDexTacWAM multi finger tactile encoderCheckpoint
Revision 17243e046b4555ca629eb7e105e31d531171c320
xinpan/roco-sharpa-pi05-rgb65-step56666Taskboard assemblyCheckpoint
Revision 0a00d4dbcd2923ab06fdd6e39eca9ea5eb5245f9
xinyuan-eval/tong-transfer-sharpatong-transfer-sharpaCheckpoint
Revision 80a25266e694e5c935ed60c435ce3068210bcfe0
yunju1115/gr00t-n1.7-h2-sharpa-cotrain-delta-ckpt79500GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 6d3255ec35d073e3a8e8e03dcaa54fa8c7af923c
yunju1115/gr00t-n1.7-h2-sharpa-cotrain-eef-camright-headonly-rel-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 896e64d527233748a7c74e85ac5f12341938999c
yunju1115/gr00t-n1.7-h2-sharpa-cotrain-refined-ckpt53500GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 99966fc6a289210f7ad7c8dd344ac8b549954196
yunju1115/gr00t-n1.7-h2-sharpa-cotrain-rel-ckpt42000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision afa484e9301f04e48c8223ca61730f6e66234082
yunju1115/gr00t-n1.7-h2-sharpa-cotrain-right-handstate-egohand03-rel-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 746104397d10809da0967be6bafe38d2fd7a2fc6
yunju1115/gr00t-n1.7-h2-sharpa-ego-eef-robotright-headonly-rel-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision e9557cb6899dcbecb31469d46b52559ce4a737b2
yunju1115/gr00t-n1.7-h2-sharpa-ego-joints-ckpt28500GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision e6af6cb81b28c695ccc5730de8a00152a4f80fd5
yunju1115/gr00t-n1.7-h2-sharpa-tele-eef-camright-headonly-rel-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 42c4fb099a0c98bf9836a2a399bb95dc077eb376
yunju1115/gr00t-n1.7-h2-sharpa-tele-right-handstate-rel-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 5bf0060a62c9304a2d83c24f44eac7838cc0dca0
yunju1115/gr00t-n1.7-h2-sharpa-tele108-ego107-0921-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision aa96fe0f7057d76e02e8bc46d5145ddd7c18adc6
yunju1115/gr00t-n1.7-h2-sharpa-tele20-from-base-joints-ckpt3500GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision ad4c014887265a67ed477fa78303a8f433368aad
yunju1115/gr00t-n1.7-h2-sharpa-tele20-from-ego-joints-ckpt4000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision db2c986cdf3dfe336d1de4f9ea1cc60d956fdfd1
yunju1115/gr00t-n1.7-h2-sharpa-tele30-0921-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 64a4f31ed425b3dcf5a0439a23f5a6d9ab2b7daa
yunju1115/gr00t-n1.7-h2-sharpa-tele30-ego107-0921-ckpt40000GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision f230f49b57fae34c90f977065e763d82267f667d
yunju1115/gr00t-n1.7-h2-sharpa-tele55-joints-0915-ckpt25500GR00T Sharpa policy variant; task coverage unverifiedCheckpoint
Revision 9602d68562776e52bd01d9b4881d41cc0fe78668
DITTO-X / DITTO-HumanBattery insertion, tong use, fragile-object handlingResearch
DexMimicGenAssembly, pouring and scene variationResearch
No matching sources.
Reviewed October 28, 2026. Variants and assets are separate entries, not separate skills. Download catalog ↓
Current status
Connecting to the collection machine…
Updated every 5 minutes. Publishing may add a short delay.
242,604Episodes saved
143,560In the model dataset
6Collection jobs running
The model dataset is a fixed pickup-and-hold subset, including validation data. New recordings enter only after their controls and task success are checked.
Collection
running
16,644 / 400,000 episodes saved
running
47,581 / 200,000 episodes saved
running
7,217 / 50,000 episodes saved
running
5,913 / 14,000 episodes saved
running
560 / 36,000 episodes saved
complete
2,005 / 4,000 episodes toward task targets
2,007 episodes saved
complete
1,612 / 1,612 recordings checked
698 episodes saved
Saved episodes passed their source’s task checks. Recent batches may still await a final audit. Select a job to see its data.
Diffusion training
System0 v2 · Object-motion model
not_started0 / 300,000 steps
Setup tests passed. The full training run has not started.
System0 v1 · First diffusion model
complete300,000 / 300,000 steps
Training finished. Simulation tests still show drops and unreliable holds.
Downloads & preparation
New source downloads
Checking…FTP-1 Sharpa recordings
runningDeCAL recordings
runningVideo-to-Data motions
runningStarting grasps
Checking…Each variant needs 128 distinct, physically checked grasps.
Early model evaluation v1 · Pickup and hold remain unreliable
An earlier v1 checkpoint, tested on two scenes. Hand-pose guidance uses a noisy robot trajectory; human retargeting was not tested.
Top left · Noisy commandsTop right · No guidanceBottom left · Hand-pose guidanceBottom right · Object-pose guidance
1× speed · Checkpoint at 230,000 steps · Two scenes, not a full benchmark.