Core service / custom collection
Build a buyer-defined first-person dataset around the tasks, environments, viewpoints, contributors, annotations, rights, and delivery format your model actually needs.
Custom collection is justified when the missing training signal cannot be licensed from existing inventory or recovered through annotation alone.
Direct answer
Custom egocentric data collection records first-person human activity to a buyer-defined specification. The specification connects a model objective to observable tasks, environments, camera placements, optional sensor streams, metadata, annotations, rights, acceptance criteria, and delivery requirements.
It is not a request for an arbitrary number of video hours. Recorded, uploaded, technically valid, task-complete, privacy-eligible, rights-eligible, annotated, and buyer-ingestible episodes are different states. The commercial unit that matters is accepted data that exposes the required signal.
This inventory was collected through licensed GIG Rewards collection programs operated with telco partners. New projects use the same evidence-first operating model while confirming project-specific contributors, environments, rights, capacity, and timing in writing.
Read the Egocentric Collection Field Guide ↗First-party collection evidence
These public derivatives show the signals EGXO has already collected across household and commercial settings. They are evidence of visible content and review practice, not a claim that a future project is already specified, synchronized, licensed, or available at scale.
Scope one environment or design deliberate transfer across both.
Choose the minimum views that keep the required action observable.
Philippines + South Africa; project availability is confirmed in the pilot scope.
10xme Technology Inc. operates EGXO Data and GIG Rewards. SIM and KYC sourcing evidence does not replace project consent.
01 / Household POV
Three reviewed phases from one maintenance recording: release a guarded component, clean the detached blade assembly, and align the guard for reassembly.
Evidence boundary: three reviewed phases from one source recording; not a continuous sequence or a complete training annotation set.
02 / Commercial Scene-first configuration
One actor-aligned POV stream is paired with frontal, side, and whole-body external views of the same work event.
Evidence boundary: manually aligned within an observed 200 ms tolerance; observed pairing only, not validated sensor synchronization.
03 / Commercial Contact-first configuration
Right- and left-wrist ego views preserve near-field manipulation while frontal and wide exo views retain body and workspace context.
Evidence boundary: visually aligned within an estimated 500 ms tolerance; not calibration-grade sensor fusion; observed pairing only, not validated sensor synchronization.
Start from the fastest path that meets the model, rights, and ingest requirements.
Use existing household or commercial data when its tasks, viewpoints, privacy status, permitted use, and delivery format already fit.
Review Available Data ↗Commission new capture for proprietary workflows, deployment environments, controlled variation, rare failures, new sensors, or buyer-defined rights and labels.
Use existing data for the baseline, locate weak task slices, then spend custom budget only on the mismatches that affect the model.
Compare Public, Existing, and Custom Data ↗The collection brief should make the required behavior, evidence, rejection rules, and final package explicit.
Target model, task, observable state, action context, outcome, failure, and evaluation question.
Task taxonomy, objects, tools, environments, starting states, edge cases, interruptions, and terminal states.
Market, eligibility, skill, language, experience, identity evidence, consent, and exclusion rules.
Head, glasses, chest, wrist, external views, RGB, audio, IMU, depth, pose, and timebase requirements.
Sessions, episodes, steps, objects, interactions, outcomes, confidence, ontology, and reviewer state.
Visibility, integrity, completeness, timing, privacy, rights, annotation, schema, and loader gates.
Permitted use, retention, vendor access, redistribution, location, transfer, deletion, and incident handling.
Manifest, schema, dataset card, checksums, lineage, known limitations, versioning, and buyer ingest evidence.
Collection environments
Household collection can target daily living, food preparation, cleaning, organization, tool use, deformable objects, long-horizon procedures, and variation across homes.
Commercial collection can target buyer-defined workplace tasks, tools, layouts, workflows, viewpoints, privacy boundaries, interruptions, and operational conditions.
Cross-environment collection is useful when the same skill must generalize across homes and workplaces. The task definition can stay stable while environments, objects, performers, and failure cases vary deliberately.
Review Household Evidence ↗Review Commercial Evidence ↗Camera placement is an observability decision. Pilot it on the real task before committing to a device or mount.
| Candidate capture | Potential strength | Critical pilot question |
|---|---|---|
| Head or glasses ego | Actor-aligned scene, attention direction, and task sequence | Do hands, contacts, and critical objects remain visible during normal head motion? |
| Chest ego | More stable torso-aligned view with useful hand and workspace coverage | Does the mount preserve near-field detail without excessive body occlusion? |
| Wrist ego | Close manipulation detail near one or both hands | What context is lost when the hand rotates, reaches, or leaves the workspace? |
| External exo | Whole-body motion, stable scene geometry, and independent context | Which contacts or object states remain occluded from the external camera? |
| Paired ego and exo | Actor detail plus body and workspace context | What clock, drift, pairing, and calibration evidence does the intended use require? |
Collection case study / commercial garment handling
The practical question was not whether four cameras could record a worker. It was which configuration preserved the required garment-handling signal, how closely the views could be associated, and whether identity redaction remained safe and precise throughout motion.
The scene-first configuration paired actor POV with frontal, side, and whole-body views. The contact-first configuration used left- and right-wrist ego views with frontal and wide external context.
Reviewers matched garment lift, contact, inspection, sorting, and placement events across the relevant views. The resulting public derivatives are visually aligned within stated tolerances, not hardware timecode synchronized.
The identity mask had to follow the actor frame by frame without assuming the head stayed in one fixed region. The earlier 960 × 540 site derivative also failed to preserve the quality of the original 1080p camera files.
The scene-first derivative was rebuilt from the four original camera files with frame-indexed head masks. The dual-wrist wide view received a dense 360-frame head track. Audio and source metadata were removed.
What passed
EGXO uses the GIG Rewards operation as a first-party collection channel. Network reach supports sourcing. It is not presented as automatic project capacity.
PhilippinesCollection market
Smart Communications · TNT · KIQ
South AfricaCollection market
Vodacom · MTN · Telkom · Cell C · SmartCall
Through GIG Rewards, the same operation reaches contributors through major telcos including Smart Communications, Vodacom, and MTN. Each collection scope confirms contributor availability, eligibility, consent, task protocol, acceptance criteria, and timing before delivery commitments are made.
Identity evidence
GIG Rewards publicly documents SIM-level identity verification, telco-authenticated contributors, and telco KYC-verified identity across its platform. That gives EGXO an authenticated sourcing channel through the same legal operator.
Network-level identity is not automatic proof that a person is eligible, available, properly informed, or authorized for a specific recording. Each custom project must define participant eligibility, notice, consent, environment permission, evidence retention, privacy handling, and any additional verification requirement.
Review GIG Rewards Identity Controls ↗Review the EGXO Contributor Rights Standard ↗Resolve visibility, operational, privacy, rights, quality, and ingest risk before volume increases.
Name the task, observable signal, deployment mismatch, intended model use, and why current datasets are insufficient.
Define environments, contributors, viewpoints, modalities, instructions, annotations, rights, quality gates, schema, and delivery.
Test difficult tasks, body types, workspaces, lighting, motion, privacy events, failure cases, and the intended loader.
Separate recorded volume from episodes that pass media, task, visibility, privacy, rights, annotation, and ingest checks.
Change the mount, instructions, task boundaries, metadata, reviewer workflow, or scope where the pilot exposes failure.
Expand only after the pilot establishes acceptance evidence, operating assumptions, release structure, and unresolved limitations.
Deliver approved assets with manifests, schema, dataset card, QA, rights, lineage, checksums, corrections, and known limitations.
Load representative episodes in the consuming pipeline and record decode, schema, modality, batch, and task-level failures.
Enterprise controls
Define the authority, lineage, permitted use, release version, transfer path, and ingest evidence alongside the media—not after collection.
Short answers for robotics and physical-AI teams evaluating a buyer-defined collection program.
Custom egocentric data collection is a buyer-defined program for recording first-person human activity around specific model objectives, tasks, environments, viewpoints, modalities, annotations, rights, quality gates, and delivery requirements. The collection should begin with a representative pilot rather than an unsupported volume commitment.
Commission custom collection when existing datasets do not match the required tasks, deployment environments, viewpoints, sensors, contributor criteria, failure cases, annotations, license, privacy controls, or delivery format. If current inventory already fits, licensing existing data can be faster and a hybrid program can reserve custom collection for measured gaps.
Yes. EGXO can scope custom collection for household, commercial, or cross-environment requirements. Exact tasks, sites, participant eligibility, privacy rules, access, permitted use, timing, and capacity are confirmed in the written project scope.
Head, glasses, chest, and wrist placements can be evaluated alongside external cameras when the model objective requires scene or body context. The pilot should prove that critical actions remain visible and that motion, occlusion, privacy, synchronization, and ergonomics are acceptable before scale.
GIG Rewards publicly documents SIM-level identity verification, telco-authenticated contributors, and telco KYC-verified identity across its platform. For each EGXO project, contributor identity evidence, eligibility, consent, availability, and any additional KYC requirement are confirmed in the written scope rather than inferred from network reach alone.
Native media with structured manifests, RLDS, LeRobot, and buyer-specific exports can be evaluated against the capture and training pipeline. The final schema and format are confirmed through a representative ingest test; a format name alone does not prove compatibility.
Pricing and timing are scoped after EGXO reviews the model objective, task mix, environments, viewpoints, contributor criteria, modalities, annotations, rights, scale, acceptance rules, security, and delivery requirements. A pilot measures the real acceptance rate and operational constraints before a larger commitment is made.
Not automatically. Human video can expose task structure, hand-object interaction, visible state changes, language, and recovery behavior, but it does not contain the target robot's joint commands, force, torque, or tactile state unless those signals are collected separately. The project should define how human demonstrations connect to robot-native data, teleoperation, simulation, or embodiment mapping.
Custom collection brief
Share the model objective, tasks, environments, viewpoints, contributors, modalities, rights, acceptance criteria, approximate scale, and target delivery path. EGXO will determine what a representative pilot must prove.
Scope Custom Collection