AI training data · human-collected
Custom video, image, speech, audio, and structured human data collection for AI and robotics teams. Tell us what you need captured, and we'll scope the people, locations, equipment, and delivery around your brief.
Custom collection · U.S. projects assessed by brief · Pilot-first
Why this is ReachIRL
ReachIRL coordinates people in real-world projects. For brands, the output can be attention and engagement. For AI and robotics teams, multimodal data collection can produce video, images, audio, observations, and structured human input captured against a defined brief.
Collection types
Scope one collection type or combine modalities around the requirements of the project. Each custom AI training data collection is defined with the buyer before fieldwork begins.
Egocentric, or first-person, video data collection records the participant’s point of view while they complete a defined activity in a real-world environment.
Household tasks · cooking · cleaning · organizing · object handling · navigation · narrated activityScope this collectionThird-person video and image data collection can cover fixed-view demonstrations, requested object angles, scene variations, or specified real-world environments.
Task demonstrations · object views · scene variations · environmentsScope this collectionSpeech and audio data collection can include read prompts, spoken commands, activity narration, or consented environmental audio under a defined protocol.
Prompts · commands · narration · agreed environmental audioScope this collectionStructured human input adds human-generated descriptions, explanations, and protocol responses to a specific task or collection brief.
Paired descriptions · action explanations · protocol responsesScope this collectionExample collection briefs
These are illustrative project types ReachIRL is prepared to scope — not claims about completed commercial datasets.
Record defined household tasks across agreed participants and environments.
Capture repeated examples of opening, sorting, assembling, placing, or manipulating specified objects.
Capture agreed routes or movement scenarios where permissions and safety requirements can be satisfied.
Record participants narrating activities or responding to buyer-specified prompts.
Managed collection
A hands-on, pilot-first process shaped around the buyer's requirements — not an automated self-serve marketplace.
Align on the activity, participants, environment, equipment, output, metadata, usage requirements, and acceptance criteria.
Run and review a small initial batch so instructions and outputs can be evaluated before expanding.
Coordinate participants, recording instructions, equipment where applicable, progress checks, and collection milestones.
Review against the agreed requirements, organize the collection, and deliver the specified files and metadata.
Defined deliverables
The exact package and controls are established during project scoping. Nothing below is implied unless it is included in the brief.
Collection package / depending on project
Collection controls / established in scope
Pilot-first
Have a task list, capture specification, or example dataset? Share the non-confidential requirements and we'll assess a pilot with defined scope, deliverables, and acceptance criteria.
Scope considers tasks · participants · locations · equipment · review · format · volume · timing
Share your briefProject inquiry
Start with whatever you know. An unusual requirement is welcome; “not sure yet” is also a valid starting point.
FAQ
ReachIRL assesses custom projects for first-person video, third-person video and images, speech and audio, and structured human input. A brief can use one collection type or combine modalities when the requirements are feasible.
Egocentric, or first-person, video is recorded from the participant’s point of view. It shows what a person sees while completing a defined activity in an agreed real-world environment.
Yes. A project can scope one collection type or combine video, images, speech, audio, and structured responses around the model and evaluation requirements.
Yes. Share the non-confidential parts of your task, capture, participant, environment, output, and acceptance requirements. We’ll identify questions and assess feasibility before proposing a scope.
Yes. A pilot-first approach lets both teams review instructions, capture quality, metadata, and acceptance criteria before considering a larger collection.
Equipment can be buyer-provided or scoped with the project depending on the capture requirements. ReachIRL does not assume a particular device or existing equipment inventory.
ReachIRL is currently assessing U.S. collection projects. Specific geographies, environments, access, permissions, participant fit, and timing are evaluated for each brief; availability is not guaranteed.
Projects can include specified session identifiers, capture settings, coverage summaries, and scoped labels or transcripts. We’ll confirm exactly what can be produced during scoping rather than assume a standard annotation package.
Pricing is scoped around the tasks, participants, locations, equipment, review requirements, delivery format, target volume, and timing.
The required participant and location permissions, incidental-person rules, sensitive-information handling, usage terms, and transfer method are established for the specific project before collection.
Custom real-world collection