Tidel AI · Humanoid + Dexterous Hand Teleoperation Dataset

What We Sell›
PRODUCT

Real-machine humanoid teleoperation episodes. Each bundles 6 synchronized camera streams, RGBD on three of them, dual-arm + dual-dex-hand commands & state, wrist 6-axis force/torque, full-body joint state, walking odometry, and VLA action-chunk annotations.
UNIT OF SALE
- One episode= one complete sub-task demonstration
- Streams6 RGB cameras + 3 depth + 11 motion topics, hardware-synced
- AnnotationsFrame-level VLA action chunks · English captions
SLICING DIMENSIONS
- By scene5 fuzzy buckets (see Catalog)
- By skill primitivedetect · grasp · lift · transport · place · retract
- By arm usageleft / right / bimanual (per-span arm field)
- By contactfree-space vs F/T-bearing (wrist wrench history)
- By locomotionstationary stance vs walking (odom delta)
DELIVERABLE FORMAT
- Per-episode treeimage_data/ + metadata.hdf5 + metadata_aligned.hdf5 + annotation/
- Open formatsJPEG · HDF5 (zstd) · JSON — no SDK lock-in
What's in the Catalog›
SCALE
- Total volume389 task hours
- Coverage5 level-1 scenes · 22 sub-tasks · 28 (scene, sub-task) pairs
- Hardwaresingle humanoid generation, biped + dex hands
- Capture~12.5 fps image base · 200 Hz motion · 6 cams sync
DISTRIBUTION (BY HOURS)
- Factory / Pipeline214 h · 55.0%
- Logistics / Warehouse109 h · 28.0%
- Service / Hospitality31 h · 8.0%
- Home / Domestic23 h · 6.0%
- Inspection / QC12 h · 3.0%
REPRESENTATIVE SUB-TASKS

Pipeline parcel sort · Bin-to-bin pick & place · Carton flap open · Bag handover · Cable / hose insertion · Tool pickup · Component fit-check · Shelf placement · Cup pour · Plate stack · Trash separation · Drawer open / close · Switch & button press
What's Inside›
CAMERAS · 6 RGB + 3 DEPTH
- image_data/sensor_camera_stereo_left_image_raw/1920×1536 · head stereo L · alignment base
- image_data/sensor_camera_stereo_right_image_raw/1920×1536 · head stereo R
- image_data/sensor_camera_fisheye_left_image_raw/1920×1536 · head fisheye L (wide FOV)
- image_data/sensor_camera_fisheye_right_image_raw/1920×1536 · head fisheye R
- image_data/sensor_camera_waist_front_rgbd_color_raw/640×360 · waist-mounted RGBD (color)
- image_data/sensor_camera_body_back_rgbd_color_raw/640×360 · body-back RGBD (color)
- image_data/sensor_camera_stereo_depth_raw/16-bit head stereo depth
- image_data/sensor_camera_waist_front_rgbd_depth_raw/16-bit waist RGBD depth
- image_data/sensor_camera_body_back_rgbd_depth_raw/16-bit body-back RGBD depth
ACTION (TELEOP COMMANDS)
- mc/left_arm/joint_command[N, 7] · 7-DoF L-arm target
- mc/right_arm/joint_command[N, 7] · 7-DoF R-arm target
- mc/left_hand/command[N, 6] · dex hand L target (thumb_swing/mcp + 4 fingers)
- mc/right_hand/command[N, 6] · dex hand R target
- mc/head/joint_command[N, 2] · yaw + pitch
- mc/waist/joint_command[N, 2] · yaw + pitch
STATE (ROBOT OBSERVATION)
- mc/joint_states[N, 28+] · full-body joint pos/vel/effort
- mc/left_hand/joint_states[N, 6] · dex hand L actual
- mc/right_hand/joint_states[N, 6] · dex hand R actual
- mc/ft_states/L_hand_ft[N, 6] · L wrist 6-axis force/torque
- mc/ft_states/R_hand_ft[N, 6] · R wrist 6-axis force/torque
- mc/imu_states/body_imubody IMU · orientation + angular vel + linear accel
- sensor/imu/orinhead-side IMU (Orin module)
- mc/leg/walking_odomodom → base_footprint · SE(3)
- /tffull TF tree across the body
SYNC · int64 ns
- base topicsensor/camera/stereo_left/image/raw (770 frames)
- align modenearest, eps=100 ms · 100% match across all 25 topics
- avg drift0–8 ms across motion + cam · 4 ms walking_odom
TASK METADATA
- annotation/<episode>.jsontask_id, src_video_name, action_chunk[] (start_frame, end_frame, English caption)
How to Evaluate›
SAMPLE PROVISION

A 1-hour sample bundle is available on access request. Sample ships in the same format as production delivery — same directory tree, same files, same annotation shape.
EVALUATION CHECKLIST · 4 dimensions
- 1. SYNC INTEGRITY25 topics aligned to stereo-left master clock; verify align_report.json per episode (avg / max drift in ms).
- 2. ANNOTATION QUALITYaction_chunk[] carries skill primitive + English caption + start/end frame. Adjacent chunks may overlap (standard VLA practice).
- 3. ACTION ↔ STATEPlot joint_command vs joint_states for any arm or hand to inspect teleop tracking; wrist F/T should spike at grasp/place.
- 4. WALKING OKIf the episode is locomotion-bearing, walking_odom SE(3) should integrate cleanly.
What It's Like to Work With Us›
ENGAGEMENT MODEL
- Point of contactDedicated per account
- CommunicationShared channel (Slack / Lark) for active customers
- Delivery digestEpisode counts and per-episode QA flags surfaced with each batch
CUSTOM COLLECTION
- Bespoke scenesAccepted on quote
- Pre-captureJoint shotlist sign-off before recording begins
- Retake loopEdge cases re-captured against agreed criteria
OPERATIONS
- Capture methodReal-machine teleoperation by trained operators
- Quality flagsPer-episode review surfaced in delivery manifest, not hidden
TYPICAL ENGAGEMENT FLOW

Sample evaluation → pilot bundle → rolling production delivery.
License Terms›
USAGE RIGHTS
- TrainingPerpetual license for model training and derivative weights
- Internal useResearch, evaluation, and product deployment permitted
- CommercialDeployment of trained models permitted
RESTRICTIONS
- RedistributionNo resale or redistribution of raw episode data
- OperatorsNo reverse-identification of human operators
EXCLUSIVITY
- StandardNon-exclusive — same episodes may be licensed to multiple buyers
- Exclusive packsAvailable on quote
Delivery & Acceptance›
FILE STRUCTURE · per episode
- image_data/JPEG per frame, 6 RGB + 3 depth folders, ns-timestamp filename
- metadata.hdf5Raw 25-topic stream (zstd-compressed records)
- metadata_aligned.hdf5All 25 topics resampled to stereo-left master clock (770 frames)
- annotation/<id>.jsonVLA action chunks (skill, frames, English caption)
- align_report.jsonPer-topic alignment QA (avg / max drift, match rate)
- topic_info.jsonPer-topic raw frame count
VOLUME
- Per episode (full)~1.3 GB at typical 60-second length (6× HD JPEG + 3× depth + HDF5)
- Per episode (color-only)~500 MB if depth folders omitted
- Catalog scaleTB-class — transfer mechanism scoped per contract
ACCEPTANCE
- Pass criteriaalign_report 100% match · annotation present · action chunks cover the active episode window
- Failed episodesRe-collection or hour-credit at customer choice

Secret access key required, provided by us upon request.

FACTORY

SERVICE

HEAD · STEREO LEFT

HEAD · STEREO RIGHT

WAIST · FRONT RGBD

HEAD · FISHEYE L

HEAD · FISHEYE R

BODY · BACK RGBD

ARMS 14 joints · rad R arm L arm

DEX HANDS 12 fingers · piano-roll

R WRIST F/T 3 force · 3 torque Fxyz Txyz

BODY head · waist · odom-z

SHIPPED WITH EPISODE RGB × 6 (head stereo · head fisheye L/R · waist · body-back) DEPTH × 3 (16-bit aligned to color) ARM joint × 14 · cmd + state DEX HAND × 12 DoF · cmd + state WRIST F/T 6-axis × 2 HEAD 2-DoF · WAIST 2-DoF FULL-BODY JOINT STATE (28+ joints) WALKING ODOM SE(3) BODY IMU + ORIN IMU TF tree 25× hardware-sync timestamps VLA action chunks · per frame

00:00 / 01:01

EPISODE

ROBOT

SCENE

INIT