Egocentric Data Collection in India: Regions, Languages and DPDP Consent

Last updat­ed: Octo­ber 2026

Key takeaways

• Ego­cen­tric data col­lec­tion India pro­grams give one team access to very dif­fer­ent kitchens, work­shops, farms, shops and ware­hous­es across regions.
• Plan cov­er­age as a grid of region x work set­ting x nar­ra­tion lan­guage, not as one nation­al quo­ta.
• Spo­ken task nar­ra­tion can be cap­tured in the par­tic­i­pan­t’s own lan­guage; set lan­guage quo­tas before recruit­ment starts.
• Indi­a’s DPDP Act and DPDP Rules 2025 shape con­sent on site, includ­ing peo­ple who appear in the back­ground of homes and shops.
• Sea­sons, fes­ti­vals, heat and pow­er sup­ply affect cap­ture sched­ules, so build them into the plan.

Robot­ics and embod­ied AI teams add India to their cap­ture plans for diver­si­ty: of set­tings, of task styles and of lan­guages. This guide cov­ers what is spe­cif­ic to ego­cen­tric data col­lec­tion in India: which regions and work set­tings to cov­er, how to plan nar­ra­tion lan­guages, which meta­da­ta to record, how to work with site own­ers, how the DPDP Act applies to head-mount­ed cap­ture, and the field real­i­ties that shape sched­ules. If you are com­par­ing ven­dors, rigs or pric­ing mod­els, read our guide to choos­ing an ego­cen­tric data col­lec­tion ven­dor first; this post does not repeat it.

Want to see India footage first? Request a sam­ple clip set.

Why do robotics teams collect egocentric data in India?

Robot­ics teams col­lect ego­cen­tric data in India because one coun­try pro­gram can cov­er a wide range of real work set­tings, from home kitchens and small work­shops to farms, shops and ware­hous­es, along with many region­al lan­guages for spo­ken nar­ra­tion. That vari­ety helps mod­els see dif­fer­ent tools, lay­outs, light­ing and task styles with­out run­ning sep­a­rate pro­grams in many coun­tries.

Research groups have drawn on Indi­an set­tings before. The Ego4D con­sor­tium, whose project site describes thou­sands of hours of dai­ly-life video record­ed in mul­ti­ple coun­tries, includ­ed a uni­ver­si­ty part­ner in India. Com­mer­cial pro­grams need more con­trol than a research bench­mark offers: cho­sen tasks, cho­sen set­tings, con­sent writ­ten for AI train­ing and an agreed spec. Our Ego4D guide explains where pub­lic data stops fit­ting.

Which regions and work settings can an India program cover?

An India pro­gram can cov­er home kitchens, small work­shops and man­u­fac­tur­ing units, farms, neigh­bor­hood shops and ware­hous­es, and each region adds its own lay­outs, tools and lan­guages. Plan cov­er­age as a grid: pick the set­tings your mod­el needs, then spread them across two or more regions so the mod­el does not learn one city’s look.

Region (exam­ples)	Work set­tings to con­sid­er	Nar­ra­tion lan­guages to plan
North (Del­hi NCR, Pun­jab, Uttar Pradesh)	Home kitchens, small work­shops, ware­hous­es near logis­tics hubs, farms	Hin­di, Pun­jabi, Indi­an Eng­lish
West (Maha­rash­tra, Gujarat)	Man­u­fac­tur­ing units, retail shops, home kitchens	Marathi, Gujarati, Hin­di
South (Kar­nata­ka, Tamil Nadu, Ker­ala, Telan­gana, Andhra Pradesh)	Elec­tron­ics and com­po­nent work­shops, kitchens, farms	Kan­na­da, Tamil, Malay­alam, Tel­ugu, Indi­an Eng­lish
East (West Ben­gal, Odisha)	Farms, mar­kets, small work­shops, home kitchens	Ben­gali, Odia, Hin­di

Treat the table as a plan­ning start­ing point, not a cen­sus. A good pro­gram con­firms access site by site, records region and set­ting in the meta­da­ta of every file, and checks cov­er­age week­ly against the grid.

Sample coverage plans for three program types

In ego­cen­tric data col­lec­tion India pro­grams, the right grid depends on what the mod­el must learn. The exam­ples below show how three com­mon pro­grams spread the same India capac­i­ty in dif­fer­ent ways. They are plan­ning pat­terns, not fixed quo­tas.

Pro­gram goal	Set­tings to weight	Region spread	Nar­ra­tion plan
Gen­er­al manip­u­la­tion pre­train­ing	Home kitchens, clean­ing and laun­dry, small repairs	Three or four regions, many house­holds per region	Free nar­ra­tion in the par­tic­i­pan­t’s lan­guage, with step names in Eng­lish
Ware­house and logis­tics tasks	Pick­ing, pack­ing, sort­ing and shelf restock­ing	Two or three logis­tics hubs, sev­er­al sites each	Short script­ed step names to align with action seg­ments
Agri­cul­ture and out­door work	Har­vest­ing, sort­ing pro­duce, tool han­dling	Regions cho­sen by crop cal­en­dar, not by city	Region­al lan­guage nar­ra­tion with a shared glos­sary of tool names

Two rules hold across all three. Keep at least two regions per set­ting, so the mod­el does not learn one city’s appli­ances or one fac­to­ry’s lay­out. And fix the share of each cell before cap­ture starts, so recruit­ment does not drift toward whichev­er sites are eas­i­est to book.

How should you plan narration languages?

Plan nar­ra­tion lan­guages by decid­ing what the mod­el will use speech for: action labels, instruc­tions or mul­ti­lin­gual ground­ing. Then set a quo­ta per lan­guage and region, agree the script and tran­scrip­tion rules, and recruit par­tic­i­pants who are com­fort­able nar­rat­ing in that lan­guage. Many par­tic­i­pants mix Eng­lish terms into their speech, so agree how code-switch­ing is tran­scribed.

• Nar­ra­tion style. Free nar­ra­tion sounds nat­ur­al; script­ed step names are eas­i­er to align with action seg­ments.
• Script. Decide native script vs roman­ized text for tran­scripts before cap­ture.
• Code-switch­ing. Mixed Hin­di-Eng­lish (Hing­lish) and sim­i­lar speech is nor­mal; tag lan­guage spans if your mod­el needs them.
• Audio qual­i­ty. Busy kitchens and work­shops are loud; set a min­i­mum audio stan­dard for nar­ra­tion.

For tran­scrip­tion con­ven­tions and dialect quo­tas, see our Indic speech data col­lec­tion guide.

Which metadata fields matter most for India programs?

For India pro­grams, the meta­da­ta that mat­ters most records where and how each clip was cap­tured: region and city, set­ting type, site type (home, own­er-run shop or fac­to­ry), nar­ra­tion lan­guage and script, ses­sion time and light­ing, and the con­sent and site-agree­ment IDs. These fields let you prove cov­er­age against the grid and rebal­ance it lat­er.

• Loca­tion: state and city or dis­trict, nev­er a street address.
• Set­ting and site type: for exam­ple home kitchen, own­er-run shop, fac­to­ry floor or farm.
• Task: task ID from your pro­to­col, plus any devi­a­tions the coor­di­na­tor not­ed.
• Lan­guage: nar­ra­tion lan­guage, script used for the tran­script and a code-switch­ing flag.
• Con­di­tions: time of day, indoor or out­door, and a sim­ple light­ing rat­ing.
• Con­sent: par­tic­i­pant con­sent ID, site agree­ment ID and the bystander steps applied.

Col­lect only what the mod­el or the audit needs. Data min­imi­sa­tion is a DPDP prin­ci­ple, so leave out per­son­al attrib­ut­es that do not serve the pro­gram, and keep iden­ti­ty data sep­a­rate from the footage.

Working with site owners in India

Most cap­ture in India hap­pens on some­one else’s premis­es, so site agree­ments mat­ter as much as par­tic­i­pant con­sent. Each site type needs a slight­ly dif­fer­ent approach.

• Homes. The par­tic­i­pant may not be the deci­sion-mak­er for the house­hold. Agree cap­ture times with the house­hold, and agree which rooms are in scope.
• Own­er-run shops and work­shops. Get the own­er’s writ­ten agree­ment, even when the par­tic­i­pant is a fam­i­ly mem­ber. Agree what hap­pens when cus­tomers walk in.
• Fac­to­ries and ware­hous­es. Expect safe­ty induc­tions, pro­tec­tive equip­ment rules and cam­era-free zones, such as areas with con­fi­den­tial designs or screens.
• Farms. The land own­er and the work­er may be dif­fer­ent peo­ple; both need to agree. Plan around har­vest win­dows and trav­el time between plots.

Record every site agree­ment with an ID and link it to the files cap­tured there. Buy­ers increas­ing­ly ask for that link dur­ing audits.

What does the DPDP Act mean for egocentric capture in India?

Indi­a’s Dig­i­tal Per­son­al Data Pro­tec­tion Act, 2023, applies when footage can iden­ti­fy peo­ple, which head-mount­ed video often does. The DPDP Rules, 2025, noti­fied in Novem­ber 2025, phase in over 18 months and require clear, stand-alone notices for con­sent. For ego­cen­tric cap­ture, that means writ­ten con­sent for AI train­ing from par­tic­i­pants, per­mis­sion from site own­ers and a plan for peo­ple in the back­ground.

On site in India, the peo­ple in the back­ground are usu­al­ly fam­i­ly mem­bers at home, co-work­ers in a work­shop or cus­tomers in a shop. A prac­ti­cal bystander plan includes sig­nage or ver­bal notice at the start of a ses­sion, a way for any­one to ask to be left out, cap­ture win­dows that avoid busy hours, and blur­ring of faces that did not con­sent. Our DPDP Act guide for AI train­ing data cov­ers roles, the rules time­line and buy­er oblig­a­tions in detail. This sec­tion is gen­er­al infor­ma­tion, not legal advice.

Plan­ning cov­er­age? Ask us for the India cov­er­age plan­ner (regions x set­tings x lan­guages).

Field realities that shape an India capture schedule

• Heat. Sum­mer heat short­ens com­fort­able ses­sion length and makes phones and cam­eras throt­tle; plan breaks and spare devices.
• Mon­soon. Rain affects out­door work, farm tasks and trav­el between sites; sched­ule out­door tasks around it.
• Fes­ti­vals and har­vests. Work pat­terns change around major fes­ti­vals and har­vest peri­ods, which can help or hurt depend­ing on the task.
• Pow­er sup­ply. Short out­ages still hap­pen in some areas; keep charged bat­ter­ies and offline stor­age on site.
• House­hold norms. Some house­holds pre­fer a coor­di­na­tor of the same gen­der, or cap­ture only at cer­tain times of day. Ask, and plan recruit­ment around it.
• Small-site per­mis­sions. Many work­shops and shops are own­er-run; get the own­er’s writ­ten agree­ment, not just the work­er’s.
• Upload band­width. Mobile data in small­er towns can be slow for large video files. Plan local back­ups and sched­uled uploads from a hub, rather than live uploads from each site.
• Device trav­el. Rigs move between cities by couri­er or with coor­di­na­tors. Track ser­i­al num­bers and cal­i­bra­tion files per device, so every clip still maps to the right cam­era.

None of these is a rea­son to avoid India. They are rea­sons to write the sched­ule with local coor­di­na­tors, who know which weeks and hours work in their region.

The BHARAT fit check for an India program

Score each item from 1 to 5 before you com­mit bud­get. Low scores point to plan­ning work, not nec­es­sar­i­ly to a dif­fer­ent coun­try.

• B: Bystander plan. Notice, opt-out and blur­ring rules for peo­ple in the back­ground are writ­ten down.
• H: Home and site access. You know which set­tings need own­er agree­ments and how they will be record­ed.
• A: Accents and lan­guages. Nar­ra­tion lan­guages, scripts and code-switch­ing rules are agreed.
• R: Regions. Cov­er­age spans at least two regions per set­ting, with meta­da­ta to prove it.
• A: Account­able con­sent records. Every file links to a par­tic­i­pant con­sent record and a site agree­ment.
• T: Tim­ing. The sched­ule accounts for heat, mon­soon and fes­ti­val peri­ods.

Once India is in the plan, use the CAPTURES check­list in our ven­dor guide to com­pare providers, and the humanoid robot train­ing data guide to place ego­cen­tric video along­side tele­op­er­a­tion and sim­u­la­tion.

How Graveiens AI runs egocentric capture in India

Graveiens AI runs ego­cen­tric video data col­lec­tion through an explic­it-con­sent work­flow: par­tic­i­pant onboard­ing, record­ing and cap­ture man­age­ment, meta­da­ta tag­ging and QA. Par­tic­i­pants and site own­ers con­sent to AI-train­ing use before any record­ing, and coor­di­na­tors brief them on the task pro­to­col and bystander rules in their own lan­guage.

Our net­work has already cap­tured 150,000+ con­sent-backed ego­cen­tric videos across Indi­a’s kitchens, work­shops, ware­hous­es, farms and stores, draw­ing on 500+ indus­tries and busi­ness­es. Nar­ra­tion and meta­da­ta can be deliv­ered in 25+ lan­guages, with deep Indic cov­er­age. Clients are invoiced only for approved deliv­er­ables, and Graveiens AI is ISO 9001 cer­ti­fied.

Frequently asked questions

Scope an India egocentric pilot

For ego­cen­tric data col­lec­tion India pilots, tell us your tasks, set­tings, regions and nar­ra­tion lan­guages, and we will come back with a cov­er­age plan. Explore ego­cen­tric video data col­lec­tion or scope a pilot billed only on approved files.

Sources

1. Min­istry of Elec­tron­ics and Infor­ma­tion Tech­nol­o­gy, Gov­ern­ment of India. The Dig­i­tal Per­son­al Data Pro­tec­tion Act, 2023. meity.gov.in
2. Press Infor­ma­tion Bureau, Gov­ern­ment of India. “Gov­ern­ment noti­fies DPDP Rules to empow­er cit­i­zens and pro­tect pri­va­cy,” 14 Novem­ber 2025. pib.gov.in
3. Ego4D con­sor­tium. Ego­cen­tric 4D Per­cep­tion (Ego4D) project site. ego4d-data.org

About the author

Graveiens AI Edi­to­r­i­al Team. The team writes from day-to-day work run­ning con­sent-backed data pro­grams for AI and robot­ics teams, includ­ing first-per­son video cap­ture, anno­ta­tion and mul­ti­lin­gual voice data.