We film the full loop with a head-mounted camera and fixed cameras at the store entrance and in the lot. The footage shows carts being pulled apart when they are stuck, turned to nest, pushed into a line, and the lean the worker uses to start a heavy train rolling. It records the strap or the hand position that keeps the train together, the route across the lot, and the stop before a car reverses.
What a researcher gets is a navigation task with a dynamic payload in a shared space. The worker constantly reads the lot: which carts to grab first, where the gap in traffic is, how many carts the slope allows.
Each retrieval run is an episode with steps like "unnest cart", "build train", "cross lot", and "return to corral", tagged for success or failure, including trains that broke apart. Hand and object tracking are optional. Files go into your cloud bucket in the exact schema you train on. Compensation and written consent are standard, and the footage carries commercial training rights.
Request a sample and it will be delivered within a business day.