Video Examples
These clips show the kind of real-world task context a partner site can help capture: people, tools, hand work, workspace layout, and site constraints.
Courtyard Orientation
What it shows: Workers stage equipment and move around an outdoor task area with long metal poles and capture hardware.
The context: This video illustrates how our trained GI operators integrate into a partner's staging area. A worker is seen holding the UMI capture rig, a specialized, gripper-style data collection device. By carrying the rig through the courtyard alongside standard scaffolding pipes, the camera and sensors record spatial awareness, worker movement, and how to navigate physical site constraints and outdoor lighting conditions.
Pipe Assembly Demo
What it shows: Workers align, handle, and fasten pipe sections while the capture rig stays close to the manipulation task.
The context: This clip highlights structured UMI data collection during a heavy, contact-rich physical task. Two workers collaborate to assemble scaffolding. One worker physically guides the gripper-style capture device to grasp, align, and manipulate metal clamps onto a pipe. This allows the rig's sensors to record the exact physics, tool handling, force, and micro-movements required for a robot to eventually assist with construction or maintenance tasks.
Balcony Ego View
What it shows: First-person capture of a railing, exterior boundary, and approach path for spatial context.
The context: Shot entirely from a first-person (Ego) perspective, a worker uses the UMI capture rig to perform a precise edge-work task: welding an outdoor metal balcony railing. This dual-capture method records both the structured movement of the gripper and the worker's natural line-of-sight. It provides AI models with critical context on how a human approaches a workspace, maintains visual focus during a high-stakes task like hot work with sparks, and operates safely along an elevated exterior boundary.
Sink Gripper Demo
What it shows: A gripper-style capture device is guided around a metal bowl and sink during a contact-rich cleaning task.
The context: This video demonstrates how the capture rig records fine motor skills for the hospitality and food service industries. The operator guides the UMI gripper through a stainless steel commercial sink, handling a metal bowl and a cleaning cloth under running water. This captures highly valuable structural data on hand-object contact, fluid interaction, and repetitive cleaning motions, teaching robotics how to navigate slippery, complex environments like a prep kitchen.
Who This Fits
This can work for restaurants, supermarkets, studios, dorms, clinics, gyms, warehouses, car shops, construction-related operations, and other sites where people perform repeatable physical tasks.
Credibility And Scale
GI already works with companies like Amazon, OpenAI, and NVIDIA. This is meant to build real supplier capacity for serious AI and robotics data buyers, not a one-off content project.
A first site pilot can be 50-150 candidate hours, and a larger validated pilot can be around 1,000 accepted hours. If a site proves reliable and a larger customer program is won, the business can scale into hundreds of thousands or millions of accepted hours across multiple environments.
Preferred Data Style
The best data is in-the-wild shift work: real skilled workers doing their normal jobs in the actual operating environment. We can also isolate specific approved tasks for repetition when that is useful, safe, and not disruptive.
What We Collect
Structured hand / tool demonstration data using capture equipment, ideally guided by the site's real skilled workers and supported by trained GI operators.
First-person video from approved skilled workers wearing a head camera while they go about normal shift work, subject to consent and site rules.
How The Site Participates
- GI can bring hardware, trained workers, and an onsite technical lead.
- The site provides access, workflow guidance, safety rules, privacy boundaries, supervision, and stop authority.
- We start in approved rooms, pre-opening / post-closing / low-traffic windows, non-critical work periods, or controlled support tasks.
- We avoid customer data, private screens, proprietary material, sensitive areas, and anything the site does not approve.
Commercial Structure
The arrangement can be structured as a flat site fee, coverage of approved incremental costs, or a share of revenue from accepted paid data hours, depending on what fits the site and the pilot scope.
Would you be open to meeting in person for lunch / coffee, or doing a phone call, so we can identify which workflows are safe, useful, and low-disruption to record?