The 50–100 hour egocentric data pilot
Pick your task categories. We pick the sites, recruit the operators, capture the data and ship audit-ready clips. East India, in weeks — not quarters.
The offer: a 50–100 hour pilot in Kolkata / Howrah. Task categories you choose. You send equipment and guidelines. We deliver consented, quality-checked data within 2 weeks.
The pilot is designed for robotics AI labs that want to test real-world egocentric data from a new region before committing to a larger programme. It is small enough to start quickly and large enough to tell you whether the data improves your model.
How it works
- 1
You choose the task categories
Assembly, cooking, packing, repair: whatever your model needs.
- 2
You send equipment and guidelines
Your cameras or sensors and your capture protocol.
- 3
We deliver within 2 weeks
Consented, quality-checked data with metadata and an audit trail.
Timeline
-
Week 1
Site onboarding + operator recruitment
We match your task categories to sites in Kolkata and Howrah, agree access with owners and managers, and sign site agreements. In parallel we recruit and brief field operators on your protocol and collect written informed consent in Bengali, Hindi or English. Your equipment is set up and tested on site.
-
Week 2
Capture + QC + delivery
Operators record your tasks under supervision. Every clip gets frame checks, metadata (operator ID, site, task, consent reference) and passes an approval gate. Approved clips are delivered by secure upload with a full audit trail.
Sample clips from the field
Two first-person clips from one packing and dispatch task, recorded head-mounted in 4K by an Orchis Green field operator in West Bengal.
What you get at the end
- 50–100 hours of egocentric footage recorded to your protocol.
- Operator ID, site, task and consent reference on every clip.
- Consent records and site agreements behind every clip. See compliance.
- A firm quote for scaling up, based on real numbers from your own tasks.
What the pilot can cover
Any of our services can be tested in a pilot: egocentric data collection, VLA training data, imitation learning data and humanoid robot data. Sites include factories and workshops in Howrah and logistics hubs, kitchens and garment clusters around Kolkata.
Questions
What is egocentric video data and why does Physical AI need it?
Egocentric video is first-person footage recorded from a head- or chest-mounted camera while a person does a real task. It captures hand–object interaction, gaze and the actor's point of view, which is close to what a robot's onboard camera sees at deployment. That makes it some of the most useful training data for Physical AI, humanoid and embodied AI models.
Why collect robot training data in East India rather than Delhi or Bengaluru?
Field operations in Eastern India cost less to run than in the crowded metros, the factories, warehouses, kitchens and workshops here are environments most models have not seen, and no major data collector works in the region yet, so sites and operators are not already committed elsewhere.
What kinds of sites and tasks can Orchis Green capture?
Fabrication, machining and assembly in Howrah and Liluah factories; picking, packing, sorting and dispatch in Dankuni and Rajarhat logistics hubs; cutting, stitching and finishing in the Bantala and Metiabruz leather and garment clusters; and everyday skilled work in Kolkata kitchens, workshops and garages. Task categories are chosen by you.
Is the data collection consent-first and compliant with India's DPDP Act?
Yes. Every participant gives written informed consent in Bengali, Hindi or English before a frame is recorded, every site is covered by a written agreement, and every clip carries an audit trail. Workflows are designed with India's Digital Personal Data Protection Act, 2023 in mind. Faces can be blurred on request, and footage is used only for AI training and never resold.
How does the 50–100 hour pilot work?
You pick the task categories and share your capture protocol and camera settings. We select the sites, recruit and manage the field operators, capture the data and ship consented, quality-checked, audit-ready clips, typically within a few weeks.
What data formats and hardware do you support?
We record to your protocol on your chosen head- or chest-mounted cameras and sensors, including 4K egocentric video, and can support on-site hardware deployment for robot evaluation runs. Deliverables include clip metadata, operator ID and secure upload.
Do you collect training data for VLA models and imitation learning?
Yes. Head- and chest-mounted first-person video of real tasks is exactly the human demonstration data that vision-language-action (VLA) models and imitation learning pipelines train on. We record to your capture protocol, camera settings and sensor set, so this VLA training data drops straight into your existing data pipeline.
Can you capture humanoid and robot manipulation data?
For humanoid robot data, we capture the human side of manipulation: hands-on tasks such as assembly, packing, cooking, sorting and repair, recorded egocentrically so the viewpoint and hand-object interaction match what a humanoid or robot arm camera will see. We also support on-site hardware deployment for your own robot evaluation runs.
How is egocentric data collection in India priced?
Every engagement is scoped as a written proposal with fees, hours and deliverables agreed before work starts. Field operations in Eastern India cost less to run than in Delhi NCR or Bengaluru, and the 50–100 hour pilot is the fastest way to get a firm quote for your task categories.
Who owns the footage and how is it licensed?
Footage is collected for your model training under a signed agreement and is never resold or reused for other clients. Every participant has given written consent, every site has a written agreement, and every clip carries the metadata and audit trail your legal team needs to sign off.