Robot training data from Gujarat's textile and garment factories
Folding a shirt, guiding a seam through a sewing machine, squaring up a saree before it's packed: cloth bends, slips and bunches, and no two pieces behave quite the same. That makes textile work hard for robots to learn. We record it in garment units and textile mills in Surat and Ahmedabad during normal shifts, and deliver egocentric video and other manufacturing data for robotics, raw or labelled.
The hand work
Most of the work in a garment unit is done by hand, even where the machines are new. Cutters spread layers of fabric on a long table and follow a marker with a straight knife. Machine operators line up two edges, feed them under the presser foot and pivot at corners with the needle down. Helpers trim threads, turn garments right side out, press them, fold them around a board, tag them and slide them into polybags.
In Surat the work is more about long lengths of fabric. Checkers unroll cloth and look for flaws, folders turn sarees and dress material into neat, even stacks, and packers bundle them for the market. Most embroidery now runs on computerised machines, but premium pieces are still embroidered by hand, and weavers walk the looms tying broken threads back in.
For a robotics team this is deformable-object data: two hands managing material that changes shape as it's handled, and the small corrections a skilled worker makes without thinking about them.
Where we record
- Surat
- Synthetic fabric, sarees and dress material. Inspecting, folding and packing, embroidery and loom work.
- Ahmedabad & Sanand
- Garment units in and around a city that has long been a cotton and denim centre. Cutting, sewing, pressing, folding and packing.
Capture setups
- Head or chest camera
- A first-person view of both hands on the cloth. Head mounts follow the eyes to the needle or the fold; chest mounts are steadier for people who stand and move around a table.
- Wrist cameras
- Close views of fingers pinching, smoothing and tucking fabric, which other cameras often can't see.
- Multi-view rig
- Two to four fixed cameras around a folding or cutting table, for repeatable angles and 3D hand pose.
- Teleoperation
- If you send a robot and its controls, we set it up at a folding or packing station and experienced workers do the task through it.
- Labels
- Step boundaries such as pick up, align, fold and bag, hand keypoints and a short description of each action, or your own scheme.
Consent and privacy
Every worker who appears in the footage signs an informed consent form, and the unit owner signs an agreement covering what we record and how you can use it.
Many garment units sew for buyers who don't want unreleased designs seen, so we agree with the owner beforehand what can be filmed. Faces, brand labels and tags can be blurred before delivery.
Start with a pilot
Tell us which tasks you need recorded, the camera setup and the labels you want. We'll reply with questions and a plan for a pilot: a few hours of footage first, so you can check the quality before committing to more.