Short answer
Footage of people working near robots comes from facilities that already run robots or cobots, usually from fixed cameras. It can support tests of how a perception system handles people in shared spaces, coarse activity and timing analysis, and exception review. It rarely shows hand detail or the robot's viewpoint, and it carries heavy privacy and workplace-notice questions.
What this asset is, and what it is not
Most pages in this section deal with footage of people doing a task for a robot to learn from. This one concerns a different asset: footage of the shared space itself, where people and machines operate within a few meters of each other. A buyer might want it to test whether a perception stack detects a person stepping into a robot's path, to study how workers move around a cobot cell, or to understand where exceptions happen.
It is not a demonstration set. A fixed camera above an aisle does not tell a model how to grasp. For that, see what human demonstration video can and cannot do. The two assets answer different questions, and a plan that mixes them up will buy the wrong one.
Who may hold it
Facilities that already run industrial robots, cobots or autonomous mobile robots may hold fixed-camera video, camera systems tied to robot cells, and logs from the robot controller or fleet software. Integrators and operators of such sites may also hold commissioning recordings. Whether any of it was retained, for how long, and under what workplace notice is something to ask, not assume.
Two published documents are worth knowing by name. ISO/TS 15066:2016 is a technical specification on safety requirements for collaborative industrial robot systems and their work environment, supplementing the industrial-robot safety standards (ISO, 2016); ISO 10218-1:2025 addresses industrial robots as partly completed machinery (ISO, 2025). A facility's footage and risk documentation may refer to them. A buyer should not read them as a statement about what any recording shows or whether any site complies, and this page draws no compliance conclusions.
What the footage can support
| Use | Why fixed-camera shared-space footage may help | Limit |
|---|---|---|
| Person detection and tracking tests | Real lighting, clutter, partial occlusion | Faces and bodies are personal data; redaction affects the test |
| Activity recognition at coarse level | Walking, carrying, standing, operating a console | Hand-level actions mostly not visible |
| Timing and flow statistics | When people enter a cell, dwell, leave | Needs reliable timestamps and cell definitions |
| Exception and stoppage review | Link a robot stop to what was visible around it | Requires aligned logs; see aligning video with system logs |
| Evaluation of perception under real conditions | Held-out sites and shifts | Selection bias if sites choose which clips to share |
All of these are perception, analytics or evaluation uses. None is training for robot control.
Keep perception evaluation separate from worker monitoring
The same footage could be used to judge a robot's perception, or to monitor individual employees. A buyer should specify the first and explicitly exclude the second: no per-person performance scoring, no identification of named individuals, no inference about attendance or behavior. Write that into the license terms and the delivery spec. It also reduces what the supplier has to disclose and to whom.
Evaluation sets should be held out by site, shift or time as your claim requires. Do not mix them back into training data; see task understanding evaluation sets for the principle.
Privacy and notice
Fixed cameras in workplaces may capture people continuously, often under a notice written for security or process monitoring rather than for AI training. Whether that notice covers licensing footage to a third party for model development is a question about the original terms and the supplier's rights, not a technical one. The European Data Protection Board's guidelines on processing of personal data through video devices are an example of regulator guidance in this area (EDPB, 2020); a buyer outside the US may be subject to frameworks like it, while suppliers on this route are US businesses. This is not legal advice.
Blurring faces or removing identifiers does not settle permissions. Redaction changes what is visible; it does not create a right to license. Redaction also changes training value, since blurred bodies and masked equipment can break the very cues a perception test needs. See redaction in demonstration video and worker consent and identifiable people.
Preparing the material
Raw continuous recordings can be mostly empty floor. Expect sampling work to find interactions, heavy review for redaction, and alignment of video clocks with robot or fleet logs. Decide whether you want continuous recordings or curated clips; the trade-offs are in continuous shift recordings vs curated clips. If the same cameras also serve process monitoring, rights may differ by purpose.
Illustrative delivery sample
This Illustrative record shows what one delivered interaction clip could carry. Values are synthetic.
# Illustrative only - not an available dataset
clip_id: cw_000412
site_id: site_021
camera: fixed_ceiling, 15 fps, view_id cam_C
robot_context: cobot_cell_3 (type recorded; no brand required)
clock_source: facility NTP; offset to robot log measured = recorded per clip
events_in_clip: [person_enters_cell, robot_slows, person_leaves_cell]
robot_log_fields: [mode, speed_state, stop_events]
redaction: faces_blurred; body_shape_preserved; screens_masked
excluded: individual identifiers, shift rosters, performance notes
permitted_use: perception evaluation; no per-person analysis
What this footage cannot support
It cannot teach manipulation. Fixed cameras seldom see hands clearly, and the robot's own sensors are not the viewpoint. It contains no humanoid action space, and a cobot's logs describe a different body doing a different task. Footage from sites chosen by the supplier may under-represent unusual events, and facilities with an incident may not share it. A cobot cell is not a bipedal humanoid's workspace either; treat it as evidence about people near machines in a particular layout, not a general model of humanoid-human interaction. Pair it with shared human-robot workflow procedures for context on intended behavior.
Questions to ask suppliers
- What types of robots operate in the area, and are controller or fleet logs retained alongside video?
- How were workers told about the cameras, and does that notice mention third-party licensing?
- How long is footage retained, and is anything older than your retention window available?
- Are camera clocks and robot logs synchronized, and how was the offset checked?
- How is footage sampled so that interactions are not buried in empty hours?
- What redaction is applied, and does it affect body shape, hands or equipment states?
- Who decides which clips are shared, and are incidents or stops included or screened out?
- Is anyone else, such as an integrator or robot vendor, able to claim rights in the footage or logs?
- Will the supplier agree that use is limited to perception evaluation and excludes analysis of individuals?
Tell SourceX what your evaluation set needs
If you need shared-space footage for perception evaluation, describe the setting, modalities and permitted uses to SourceX. SourceX looks for US businesses that may hold such material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. A request does not guarantee a match; SourceX reviews the requirement and follows up about potential sources. See the humanoid training data guide and the workplace task demonstration hub for the rest of the section. The catalog page on video recordings covers the general asset type.
Tell SourceX what your evaluation set needs to cover
Sources
- ISO/TS 15066:2016 Robots and robotic devices — Collaborative robots — technical specification on collaborative industrial robot systems; named for orientation only.
- ISO 10218-1:2025 Robotics — Safety requirements — Part 1: Industrial robots — industrial robots as partly completed machinery.
- Guidelines 3/2019 on processing of personal data through video devices (Version 2.0) — example of regulator guidance on video devices.