Data Operations
Robot Failure and Recovery Data Collection Services
A policy trained only on successful demonstrations has never seen what going wrong looks like. The first time a grasp slips, an object lands in
November 13, 2026
Data service 02
human demonstration services
Egocentric video, hand pose, gaze tracking, and force gloves for VLA pre-training.
50,000+
Hours collected
1,200
Scene types
40Hz
Hand pose tracking
What is human demonstration?
Human demonstration is the process of a trained operator physically performing a task — grasping, pouring, assembling, navigating — while wearable sensors capture hand pose, gaze direction, body motion, and egocentric video. The resulting dataset teaches a policy how a human actually solves the task, not how a script approximates it.
Why teams partner with us
Collecting demonstration data in-house means recruiting skilled operators, sourcing sensor rigs, building a logging pipeline, and running QA across thousands of episodes. We compress that into a managed program.
- Trained demonstrators — operators calibrated on your task before a single episode records
- Full sensor stack — eye tracking, force gloves, IMUs, and head-mounted RGB-D
- 1,200+ scene types — diverse environments so your policy generalizes
- Episode-level QA — every demo validated against your acceptance criteria
Why outsource demonstrations?
Your research team should be training models, not managing capture sessions. We handle recruitment, hardware, and quality so you get clean episodes on schedule.
50,000+ hours collected to date.
40 Hz hand pose tracking across all rigs.
3 continents of operator coverage.
Where we collect
41+ delivery centers across 12 countries. Every program runs from a Roborax hub near your target time zone.
Asia Pacific
India · Philippines
Americas
USA · Canada · Colombia · Jamaica · El Salvador · Belize
EMEA
UK · Albania · Kosovo · Morocco
What we deliver
The modalities VLA models actually need
Egocentric capture across hands, gaze, scene, and force — synchronized into a single timeline.
Egocentric video
First-person scene capture at 60fps, calibrated for VLA pre-training.
Hand pose + grasps
Per-finger joint angles plus grasp state, at 40Hz.
Eye gaze tracking
Operator gaze fused with scene video for attention modeling.
Force + tactile
Glove-based force data for contact-rich tasks.
How we work
Scene design to packaged dataset
A four-stage pipeline that lands clean multi-modal data in your bucket every week.
1Step 1
Scene design
Spec the task domain and the demonstration patterns. Coverage matrix locked.
2Step 2
Operator briefing
Train demonstrators on the capture protocol. Per-task acceptance criteria reviewed.
3Step 3
Capture
Synchronized multi-modal recording. On-rig validation flags drift.
4Step 4
Label and package
Annotated, time-aligned dataset delivered as bag files or your custom format.
Rigs and tools
Glasses, gloves, and capture rigs
Best-in-class hardware for each modality.
Meta Aria
Research glasses
Quest Pro
Hand tracking
Manus VR
Per-finger gloves
Tobii Pro 3
Gaze glasses
GoPro Hero
Egocentric video
Multi-camera rig
3rd-person sync
What our partners say
Their demonstration pipeline gave us fifty hours of clean egocentric data with hand pose in three weeks. That was enough to bootstrap our VLA pre-training from scratch.
Lin Park
Research Director, Manta Labs
FAQ
Questions about human demonstration data
What is the difference between teleoperation and human demonstration?
Teleoperation uses a control interface — the operator never touches the robot. Human demonstration captures natural human motion directly, which is then used to train imitation learning policies. Both produce trajectory data but from different sources.
How natural are the demonstrations?
Operators are briefed on task objectives but not scripted on exact movements. This produces natural variation in approach and execution — which is exactly what imitation learning needs to generalize.
What body types and skill levels do operators have?
Our operator network spans diverse demographics, hand sizes, grip strengths, and experience levels. If your task requires a specific operator profile — handedness, physical dimensions, domain expertise — we can filter for it.
How do you ensure consistency across operators?
Every demonstration is reviewed against a task specification before acceptance. Outliers are flagged and sent back for re-capture. Inter-operator consistency metrics are included in every delivery report.
Further reading
From the blog
VR Teleop vs. Physical DemonstrationWhich method produces better training data and when.
From the blog
Teleop Operator Fatigue: The Hidden VariableHow fatigue affects data quality and what to do about it.
Build your demonstration pipeline
Specify the scene domain and modality mix. We scope a sized program in two days.
FROM THE FIELD
Data collection insights
Robot Training Data
Edge-Case Data Collection Services for Robotics AI
Most robots look great in the demo. The trouble starts in the field, where a reflective surface, an unusual object or a half-second sensor dropout
November 12, 2026
Multimodal Sensor Capture
Outsourcing Robotic Sensor Data Collection for AI Research
Research teams rarely lack ideas. They lack hours. Every month spent building rigs, calibrating cameras and recruiting operators is a month not spent on the
November 11, 2026
Teleoperation Data
How to Choose a Robot Teleoperation Data Collection Partner
Hiring a data vendor feels like a procurement decision until the first batch arrives. Then it becomes a modelling problem. The operators your partner employs,
November 10, 2026
Robot Training Data
Generative Video and Gaussian Splatting for Robot Data Augmentation
Collecting a real demonstration costs operator hours, hardware time and a scene. Synthetic data augmentation robotics teams now use can multiply what one demonstration is
November 9, 2026
Data Operations
Mixing Real and Synthetic Data: Finding the Right Ratio for Robot Policy Training
Every robotics team eventually asks the same question: how much simulation is too much? The honest answer is that there is no universal real and
November 6, 2026
Explore more services
Seven services. One synchronized pipeline.
Teleoperation
VR and leader-follower robot control logging.
Multimodal sensor capture
RGB-D, LiDAR, force, and tactile streams.
Annotation and labeling
Bounding boxes, segmentation, action labels.
Synthetic and sim-to-real
Domain-randomized scenes and sim transfers.
Evaluation benchmarks
Held-out test sets and success-rate scoring.
Long-tail and edge-case capture
Rare scenarios your policy will face in production.