Egocentric Data Collection in Mexico

Egocentric data is video recorded from the point of view of the person doing the work, with their hands, tools and the objects they handle in frame. Commercial egocentric data collection narrows that to paid, real-world work settings: the shelves, stockrooms, service bays, commercial kitchens and workbenches where the tasks an embodied AI or robotics model has to learn actually happen.

In Mexico, Firsthand runs this as custom collection. You write down what the model is missing, which tasks, which environments, which conditions and how deep the annotation goes, and contributors in Mexico are recruited to capture exactly that, in the commercial settings the spec names.

Geography is a diversity axis in its own right. Packaging, tools, store layouts and work habits differ from market to market, and a model trained on one country's footage quietly overfits to it. Adding Mexico to a spec is how a program widens that spread on purpose instead of collapsing onto whoever was easiest to recruit.

Mexico is one of the jurisdictions with depth in Firsthand's published coverage data: a named legal basis (LFPDPPP) is on file, and one commercial environment is already listed for it.

Listed environments are where a spec can usually start fastest. Other commercial environments are still recruited to spec when a program asks for them.

"Listed" means the environment already appears for Mexicoin the published coverage data. "Recruited to spec" means contributors are sourced for that environment when a spec asks for it; pool sizes are quoted per engagement rather than published.

Household chores and home routines are excluded from this program. If your model needs home footage, see everyday routine collection instead.

The default tier for commercial collection in Mexicois a recent-generation smartphone worn in a chest mount. It keeps both hands free and in frame, it is quick to deploy on a shop floor, and it records the task from roughly where the worker's eyes are. The project sets the minimum device spec, and contributors whose device does not meet it are not admitted to that project.

When a spec needs more than phone video, the same program can step up a tier: a 360 head mount for wider context, or the full instrumented rig with a head-mounted camera, dual wrist cameras, depth and IMU, synchronized to a shared clock. Same spec, same QA bar, whichever tier the project uses.

See egocentric video data collection for how the tiers compare, and sample data outputs for the files a recording ships with.

Mexico is one of the jurisdictions whose consent framework is already documented in Firsthand's provenance section, so a named legal basis can be printed: LFPDPPP. The official regulator or statute page is linked in the sources below.

Releases are collected before capture, and contributors stop recording when an un-consented person enters the manipulation region. Every hour ships with its consent record, so the provenance of a clip can be traced back to a signed release.

A documented basis is a starting point, not a substitute for the engagement paperwork: environment access and site permission are still confirmed for every location a contributor proposes.

Read the full chain of custody on the provenance page.

Commercial settings put bystanders, customers, vehicles and screens in frame far more often than a home does. Every clip runs through the anonymization pipeline before delivery: faces, licence plates and screens are detected, tracked across frames, redacted and then checked in a human review step.

Detection works on bounding boxes, so what gets blurred is auditable rather than a black box. The method is documented on the anonymization page.

Footage from Mexico is delivered in the same formats as every other Firsthand program: RLDS, LeRobot, WebDataset, HDF5, zarr and Rerun .rrd, each mapped to the canonical episode schema. Annotation depth, from action segments to hand pose, is whatever the spec asked for.

Data ships under a buyer-owned commercial licence. See delivery formats for the schema and pricing for how a quote is built.