High-quality STEM problems
Original master's-level and above problems in maths, physics, chemistry and biology, each with a detailed worked solution.
6,000Problems, split by subject on request
Training data from the physical world.
Collov Data turns real expert work, real spaces and real robots into datasets frontier labs can train on today — expert-built post-training sets, parsed interiors and 2.3 million real-robot episodes.
Unretouched first frames from delivered episodes. Every frame maps to a specific episode, robot serial number, capture configuration and MCAP recording you can replay before you buy.
Model training data10 datasets
Off-the-shelf reasoning, code, agentic, expert-domain and pretraining sets for language and multimodal models.
See the catalogSpatial dataReal interiorsParsed scenes, multi-step edit trajectories and spatial evals from Collov's production visual-agent pipeline.
See spatial dataPhysical AI data2,341,350Physics-verified real-robot manipulation episodes across 8 commercial domains and 34 robot platforms.
See robot dataModel training data
Expert-built datasets for frontier post-training.
Original, expert-written data for reasoning, code, agents and professional domains — ready to ship today and already delivered to tier-one model labs. Download a sample before you talk to us.
10Datasets ready to ship
9Original, expert-written
HLE-levelDifficulty on flagship sets
5With downloadable samples
Multi-domain expert questions
Medicine, finance, law and computer science at senior-undergraduate to PhD level, with professional answers to real-world problems.
5,000Items, with engineering coming next
Hard cybersecurity RL environments
Expert-built challenges with Docker environments, attachments, verifiable grading and an oracle answer for every item.
UnsolvedBy current frontier models, every challenge
Full catalog
Open a row for details and a sample.
Original problems in mathematics, physics, chemistry and biology at master's level and above, each with a detailed worked solution. Built for deep-reasoning training.
View sample dataVolume6,000 problems
DifficultyMaster's / PhD
Quality and track recordHLE-level difficulty; delivered to multiple tier-one labs
NotesCan be split by subject
High-quality questions from medicine, finance, law and computer science, from senior undergraduate to master's and PhD level. Covers real-world problems with professional answers to strengthen understanding and generation in complex professional domains.
View sample dataVolume5,000 items
DifficultyMaster's / PhD
Quality and track recordHLE-level difficulty; delivered to multiple tier-one labs
NotesEngineering and further disciplines coming
Detailed diagnoses of the logical and factual errors in model outputs. Trains a model to review its own work — spotting and fixing factual errors, contradictions and reasoning leaps — for more rigorous, reliable answers.
View sample dataVolumeSupplied on demand
Difficulty—
Quality and track record>95% accuracy; delivered to a tier-one lab
NotesCustom error types supported
Original competitive-programming problems with sample input and output, editorial solutions and time and space complexity analysis. Covers classic algorithms and data structures for ICPC, LeetCode and Codeforces-style problem solving.
View sample dataVolumeCustom volume
DifficultyTop competition level
Quality and track recordBenchmarked to ICPC / CCPC
NotesCan be split by difficulty or problem type
Pretraining corpora for a range of needs, covering text, images and video.
Request a sampleVolumeCustom volume
Difficulty—
Quality and track recordDelivered to multiple tier-one labs
Notes—
Original illustrated problems in physics, chemistry and biology from senior undergraduate to master's level and above, with detailed worked solutions, for multimodal deep-reasoning training.
View sample dataVolumeCustom volume
DifficultySenior undergrad to PhD
Quality and track recordDelivered to multiple tier-one labs
Notes—
Agentic task samples drawn from real, high-value knowledge work. Each includes a task description, the related files and grading rubrics.
Request a sampleVolumeCustom volume
Difficulty—
Quality and track recordDelivered to a tier-one lab
Notes—
Multi-turn instructions from real overseas users to build, revise and maintain high-quality websites over time, each reviewed and scored by design experts. Suited to high-quality trajectory generation and reinforcement learning for code generation and agent behaviour.
Request a sampleVolumeCustom volume
Difficulty—
Quality and track recordExpert-scored; delivered to a tier-one lab
Notes—
Built by security professionals to improve model cyber capability. Each item includes the challenge, required environment and attachments, the Docker environments needed to solve it, a verifiable evaluation method and an oracle answer. Every challenge is currently unsolved by frontier models.
Request a sampleVolumeCustom volume
DifficultyBeyond current models
Quality and track recordVerifiable evaluation; delivered to a tier-one lab
Notes—
High-quality questions in medicine, finance and law with point-by-point grading rubrics, from senior undergraduate to master's and PhD level. Covers real-world questions and professional scoring points to improve answers in these domains.
Request a sampleVolumeCustom volume
DifficultySenior undergrad to PhD
Quality and track recordDelivered to a tier-one lab
Notes—
Spatial data
The spatial data engine.
Every run of Collov's visual agent produces structured spatial data as a by-product. Collov Data packages it for frontier labs, robotics teams and enterprises whose AI has to work indoors — with people reviewing along the way.
The loop that runs Collov's products — perceive the scene, plan, act, check the result, learn — writes a structured record at every pass. That record is the dataset.
Parsed scenes
Real interiors broken down into objects, materials, geometry and the spatial relationships between them — structured, queryable scene state rather than flat labels.
Edit trajectories
Complete multi-step edit histories from production traffic: what was asked, what changed, what was corrected. Not synthetic one-shot before/after pairs.
Spatial evals
Benchmarks that check whether a model respects real-world constraints — depth, planes, occlusion, scale and placement — built on the same approach as Collov's Spatial Arena.
Humans stay in the loop: reviewers check and correct agent output before it becomes training data.
Physical AI data
Real-robot manipulation data at foundation-model scale.
A production data supply, not a research dump. Licensed, deduplicated and physics-verified episodes captured on real hardware, shipped as MCAP and LeRobot v3.0 — and expandable to your task list on a six-week capture cycle.
2,341,350Episodes, complete and reviewed
44,000+Recorded hours, multi-stream
1,240Annotated tasks in 450 scenarios
34Robot platforms from 11 OEMs
2.9 PBRaw corpus, compressed MCAP
Why internet video can't train a robot.
Compute scaled. Architectures scaled. Physical interaction data did not. Passive footage is missing the four things a manipulation policy needs most.
No contact forces
Video shows what moved, never the force that moved it. Grasp stability can't be learned from pixels alone.
No action labels
No joint targets, end-effector commands or gripper state — nothing for a policy to imitate.
No physical consistency
Annotators can't check momentum or kinematic limits, so noise and teleop artefacts slip into training.
No embodiment transfer
Single-robot corpora overfit one kinematic chain and fail on a new body.
We record the full causal loop.
The robot's own state at up to 1 kHz, the commanded action, the resulting contact and the sensed outcome — time-synchronised in one MCAP episode.
- 1State
- 2Intervention
- 3Outcome
- 4Feedback
94%Task success on verified data
52%Same model on raw video
Eight commercial domains, collected to production depth.
Every domain below is live, delivered volume — not a roadmap. Select one to see its scenarios.
Clothing foldingHome service robotics
The deepest domestic manipulation corpus in the catalog: kitchens, bathrooms, bedrooms, living rooms and entryways, captured in real furnished apartments rather than staged rigs.
759,480Episodes
250Scenarios
720Tasks
32.4%Of corpus
BimanualLong-horizonArticulated objectsDeformables
Delivered episodes by scenario
Clothing folding and storage202,910
Tabletop item tidying195,470
Dry trash cleanup77,920
Scattered item sorting and storage68,970
Office supply organization49,190
Bathroom sink tidying45,660
Top 17 of 250 scenarios. 233 more in the library, including kitchen cleaning, cooktop cleaning, mixed-clutter sorting and food handling.
Thirty-four platforms from eleven OEMs.
Wheeled mobile manipulators, full-size humanoids and force-controlled arms — the range of bodies a generalist policy needs, in one licence.
Delivered hours, top seven platforms
Astribot S127,000 h
AI² Alphabot 27,430 h
Neomatrix L14,070 h
UBTech Walker S22,430 h
Neomatrix Rizon 4s1,650 h
Neomatrix M11,140 h
Unitree G1850 h
±0.1 mmArm repeatability (Astribot S1)
>10 m/sEnd-effector max velocity
15 kgDual-arm payload (Walker S2)
275 TOPSOn-board compute (Walker S2)
Fleet composition
AstribotWheeled mobile, 23 DOF
3 platformsNeomatrixMobile and force-controlled
6 platformsUnitreeHumanoid, 29 DOF
4 platformsUBTechHumanoid, 30 DOF
3 platformsAI² RoboticsWheeled dual-arm
2 platformsFive further OEMsHumanoid and dual-arm
16 platformsParallel grippers, five-finger dexterous hands and vacuum tooling, with 6-axis wrist force/torque sensing.
Signal density
Not a video with a caption. A multi-stream recording.
Every episode is a time-synchronised, schema-typed recording you can replay topic by topic — vision, action and force aligned to one episode clock.
Native sample rates, as recorded
Low-level actuator state1,042 Hz
Joint / proprioception250 Hz
TF transform tree209 Hz
Joint and Cartesian command100 Hz
Dexterous-hand state95 Hz
Teleop FSM command50 Hz
RGB / depth cameras30 Hz
Rates are as recorded on production lines, not resampled.
400+ recorded topic types
Camera43
TF transform tree40
Arm38
Control / teleop34
Head33
Torso33
Chassis31
Gripper29
Dexterous hand15
Joint / actuator8
Metadata7
Vacuum, IMU, leg, knee7
RGB and depth
Head, chest, torso, wrist and gripper cameras.
Proprioception
Joint position, velocity, current and command.
Force and torque
6-axis wrist F/T and end-effector wrench.
Frames and metadata
TF tree, camera intrinsics and extrinsics, SOP.
Labelled to the step, not the clip.
Every task breaks down into an ordered sequence of steps, and every step carries skill verbs from a controlled 140-term taxonomy.
Worked example: kitchen cleanup and dish storage
Unitree G1, bimanual, stereo and monocular 720p, BrainCo hands. 22 annotated steps and 13 skill verbs in total; first eight shown.
- Move the bipedal robot to the sinkMove
- Grasp the kettle with the right armGrasp
- Carry and place the kettle at its home positionCarry, Place
- Bend and collect used tissues from the counterBend, Grasp
- Wipe standing water from the work surfaceWipe
- Open the dishwasher doorOpen
- Place dishes into the rack slots, repeating per itemPlace, Repeat
- Close the door and return to the home poseClose, Adjust
From the 140-verb skill taxonomy
MovePlaceGraspCarryGripCloseWipeArrangePushTurnPick upPlace inAdjust postureClampCrouchAlignCoordinateInsertPullHandoffPressPourFoldScoop+ 116 more
Ordered step sequenceIncluded
Per-step skill verbsIncluded
Device serial and capture configIncluded
Bilingual English and Chinese aliasesIncluded
Human egocentric capture, as a complement.
Wearable first-person recordings of real human work — the cheapest route to task diversity. 40 scenarios, 1,400 recorded hours, nine rigs in rotation.
640 h
Household
Washstand cleaning, cucumber cutting, dish washing, after-meal tidying, wok stir-frying.
310 h
Factory
Boxing, weighing, gluing and sealing at a packing station; manual pallet-jack transport.
250 h
Hotel
Guest-room housekeeping: linen changes, fixture wiping, amenity arrangement.
200 h
Office and outdoor
Restroom cleaning in stereo; a six-camera outdoor waste-station rig.
RoboPocket
Compact chest or head monocular rig for long household sessions.
MRR
Chest-mounted stereo pair with dual-video output per session.
WE-Ego
Wide-FOV fisheye headset for cluttered interior work.
DAS-Ego
Six-camera cart-mounted array for outdoor and multi-worker scenes.
Our inverse-dynamics model recovers 6-DoF end-effector trajectories and contact points from this footage, turning passive video into robot-executable action labels.
Physics verification, not human eyeballs.
The EdgeWAM engine scores every trajectory against rigid-body physics in both directions before it enters a delivery batch.
Forward dynamics: consistency scoring
Does this episode obey physics?
Trajectories are checked against rigid-body kinematics, collision boundaries, joint limits and acceleration and jerk envelopes. Sensor dropouts, tracking jumps and teleoperator noise are flagged and quarantined before training — not discovered three epochs into your run.
Inverse dynamics: latent action recovery
What action produced this frame?
Recovers 6-DoF end-effector trajectory distributions and contact points directly from passive egocentric video, converting unlabelled archives into structured, robot-executable actions at a fraction of teleop cost.
| Metric | Benchmark target | Validated against | Status |
|---|---|---|---|
| Kinematic trajectory accuracy | MAE ≤ 3.1 mm / ≤ 2.8° | OptiTrack optical ground truth | Verified |
| Physical anomaly detection | 94.2% precision / 91.8% recall | Kinematic-limit and jerk injection | Verified |
| Contact force estimation | MAE ≤ 1.2 N (0.5–10 N window) | 6-axis F/T load-cell fixtures | Verified |
| Downstream sample efficiency | 38% fewer training hours | Policy convergence vs. raw data | Verified |
What it does to your policy.
Same architecture, same compute budget, same task suite. Only the training corpus changes.
Task success rate after convergence
Verified tokens (ours)94%
Filtered teleop stream78%
Unfiltered raw video52%
Semantic-only annotation41%
+42 pts
over unfiltered raw video. Physics-verified tokens remove the non-physical demonstrations a policy would otherwise learn to imitate — the largest single source of brittle failure at deployment.
38%Fewer training hours to convergence
100%Of episodes physics-scored
Delivery and licensing
Lands in your pipeline in the format it already reads.
Slice the robot catalog by platform, domain, scenario, skill or capture configuration — or take a pre-cut dataset off the shelf.
Example dataset card
Unitree G1 diverse manipulation
250 tasks across 90 home and retail scenarios — cleaning, tidying, clothing handling, refrigerator zoning, basket shopping and shelf restocking. Six capture configurations span Dex1 and BrainCo end effectors with monocular and stereo 720p vision.
8,620Records
328 hDuration
4.8 TBSize
MCAPFormat
Largest scenarios
Supermarket shelf handling85.3 h2.7k rec
Bedroom tidying71.6 h1.6k rec
Bathroom tidying and laundry prep71.1 h1.3k rec
Kitchen food handling23.8 h710 rec
Living room22.8 h590 rec
Formats
- MCAP — lossless multi-stream, schema-typed, replayable
- LeRobot v3.0 — drop-in for Hugging Face training loops
- OWAMcap — world-model token stream
- Parquet or webdataset shards on request
Slice by
- Robot platform, brand or end effector
- Domain, scenario, task or skill verb
- Capture configuration: mono, stereo, RGB-D
- Duration, success flag or object class
Step 1
Ingestion
Standardised capture in MCAP and LeRobot v3.0, with device serial, capture config and camera calibration bound to every record.
Step 2
Sanitisation
Automated GDPR/CCPA facial and biometric scrubbing, with signed commercial waivers from every operator and location.
Step 3
Verification
EdgeWAM physics scoring, duplicate detection and a causal metadata graph for every episode.
Step 4
Deployment
S3/GCS transfer or physical shipment under a tier-1 SLA. You pay only for frames that clear the agreed thresholds.
Every licence includes
- Perpetual, worldwide commercial training rights
- Full provenance chain: operator waiver, robot serial, episode
- A personal-data scrub certificate with each delivery batch
- A replayable MCAP sample set before any commitment
New capture, on your task list
Monthly robot-hours of net-new capture
21,700 hToday
52,000 hIn 12 months, contracted
Today's run-rate by line
9,600 hWheeled manipulator lines
7,900 hHumanoid lines
4,200 hForce-control and bench lines
How a commission runs — first verified delivery in six weeks
Week 0Task specification and SOP design with your team
Weeks 1–2Rig configuration, calibration and a pilot batch
Week 3Pilot review: you sign off the SOP and QA gate
Weeks 4–6Production capture at full line rate
Week 6First verified delivery, then weekly increments
Take the sample set, then take the catalog.
Tell us the embodiment, the skills or the model capability you're training. We'll cut a free evaluation slice against your task list — full annotation, no commitment.
FreeEvaluation slice
48 hoursTurnaround
6 weeksTo first commissioned delivery
