Robotics integrator video: sim render vs. real line
Plant engineers discount polished automation reels. The sim-and-real-line pair is the proof format that survives them, and cycle-time numbers wait for FAT.
The best-performing video an automation integrator can publish is almost never the drone shot of the plant. It is a forty-second split screen: the offline simulation of a cell on the left, the same cell running on the real line on the right, same camera position, same clock, three consecutive cycles, no cuts.
That format works because it answers the question the buyer is actually asking. A manufacturing engineer evaluating an integrator is not trying to find out whether you own robots. They are trying to find out whether your simulation predicts your build — because if it does, the quote you handed them means something, and if it does not, they are about to eat a schedule slip during commissioning.
The other reason to build this format: the simulation already exists. It was produced during quoting, in whatever offline programming environment your controls team uses. Turning that render into a marketing asset is a re-export, not a new project.
Why the pair survives a plant engineer
Every buyer in this category has been sold a cell that missed its numbers. The scepticism is earned, and it makes them read video differently from a normal audience. In the first ten seconds they are checking three things:
- Does the sim match the real motion? Same approach vectors, same wrist orientation, same dwell at the pick. Mismatch means the sim was decorative.
- Did every part get picked? A hero take with one clean cycle tells them nothing. Three uncut cycles tells them the gripper and the presentation actually work.
- Is anything hidden? A cut mid-cycle, a speed ramp, a camera move at the moment of the pick. Any of those and the whole library gets discounted.
A montage with music answers none of that. It is also indistinguishable from every competitor's montage, because all of them licence the same catalogue of orange arms welding in slow motion.
The pairing spec
Shoot and render to a fixed spec so the pairs are comparable across cells and across years. This is the part worth writing down once.
| Element | Simulation pane | Real pane | Rule |
|---|---|---|---|
| Camera | Virtual camera at operator eye height, outside the fence line | Tripod, same height, same standoff, locked off | If you cannot match position, do not publish the pair |
| Length | Three consecutive cycles minimum | Same | No cut inside a cycle, ever |
| Start trigger | Part-present sensor going true | Same event | Both panes start on the same signal, not on the clapper |
| Speed | 1x | 1x | If you ramp, burn the multiplier on screen and apply it to both panes |
| Audio | Silent | Real plant audio or silent | Never score the real pane with library music |
| Overlay | Running timer, cycle counter | Same timer, same counter | Typed overlays, not rendered text |
Frame rate is the detail that quietly ruins these. If the plant camera captures at 30 fps and the simulation exports at 24, the two panes drift a little further apart every cycle, and by the third one an engineer is watching the drift instead of the cell. Conform both to a single rate before you build the timeline. Versely timelines run at 25 fps by default, and dropping 24 fps clips onto a 25 fps timeline covers what conforming actually does to the footage.
Assemble it as a two-up rather than a cutaway; picture-in-picture handles the inset variant. Build the layout once and treat it as a template — see reusable editor drafts for keeping one timeline and swapping the footage per cell.
Cycle-time numbers wait for a witnessed FAT
Here is the rule worth being inflexible about: a throughput or cycle-time figure does not go on screen until a customer's engineer has stood next to the cell during a factory acceptance test and signed the result.
Not because a pre-FAT number is dishonest, but because of where it ends up. A figure that appears in a sixty-second clip gets pasted into a capability deck, then into an RFQ response, then into a scope document, and at that point it is a commitment with liquidated damages attached to it. Video is the least controlled surface you publish on and the numbers on it travel furthest.
When a number does clear FAT, it travels with its conditions or it does not travel at all:
- The part. Part number or part class, mass, and presentation — bulk, dunnage, or fixtured.
- The end effector. Gripper type and the tool changer, if there is one.
- The measurement. How many consecutive cycles, measured how, from which trigger to which trigger.
- What is excluded. Changeover, reject handling, operator intervention, and whether the number is a cycle time or an effective rate net of availability.
- The witness. "Witnessed at FAT, [month]" on screen, with the customer named only if they have agreed in writing.
Before FAT, the honest version is still publishable and still persuasive: Simulated cycle. Not yet validated on hardware. Burn that into the frame rather than putting it in the caption, because the caption does not survive a repost. Spec sheet claims that do not survive testing covers the general failure pattern.
Generate the context, never the cell
The cell itself — the robot, the gripper, the part, the fixture, the guarding — is filmed or rendered from your own simulation. There is no third option.
Generative video models produce motion that looks plausible and is physically wrong in ways this audience reads instantly: an arm with an extra axis, a gripper that closes through the part, a wrist that rotates past its limit. The version that actually damages you is a safety one — a person standing inside the envelope of a fenceless cell, a light curtain that is decorative, a scene with no visible e-stop. Publish that and you have advertised that your team does not think about guarding.
What generation is genuinely good for here is everything around the cell:
| Element | Source |
|---|---|
| Cell in motion, part, gripper, guarding | Filmed, or exported from your offline programming environment |
| Sector-flavour establishing shots, plant exteriors, seasonal context | Generated b-roll |
| Title cards, callout labels, spec plates | Typed overlays — never let a model render a number |
| Narration between beats | Synthetic voiceover is fine for a technical walkthrough |
| Captions | Burned in, because this plays silently on a phone in a plant |
Where generated footage appears in an otherwise-documentary piece, label it — synthetic media disclosure covers the mechanics, and the b-roll generator covers the job that recurs most.
One more constraint that bites integrators specifically: the part is usually the customer's intellectual property, and some of it is export-controlled. Substitute a dummy part, blur the geometry, or shoot a different cell. A badge and a plant tour are not a release — get written sign-off from the customer's engineering contact before anything filmed inside their facility goes public.
One cell, six assets
Shoot the FAT once and cut it out to the surfaces where an integrator actually gets read:
- 40-second paired proof clip. LinkedIn and sales email. This is the flagship.
- 15-second vertical cut of the real pane only. Nobody reads a split screen on a phone.
- Two-to-three minute technical walkthrough. Attached to RFQ responses, narrated, with the conditions on screen. The structure in explainer video production for complex B2B applies almost directly.
- Stills for the capability deck. Pulled from the same footage, so the deck and the video agree.
- A "what changed between sim and build" clip. The single most credible asset an integrator can publish, and virtually nobody does it. Show the gripper revision, explain why, show the corrected cycle.
- A short cell-family explainer. Palletising, machine tending, vision inspection — one per capability, sized for the feed.
Iterate the edit on 480p previews, which are free and carry a short per-user cooldown, then pay once for the final export regardless of how many clips are on the timeline — previewing before you pay for the export walks through the loop. Nothing carries a watermark on any plan, which matters when the file is going into a customer's procurement folder.
FAQ
Can we publish the simulation before the cell exists?
Yes, and it is one of the better assets you have during a long sales cycle. Label it as simulation in the frame, do not attach a validated cycle time to it, and describe it as what it is: the model your quote was built on. Publishing the sim and then publishing the paired version after FAT is a two-part story that demonstrates exactly the thing you are selling.
The customer will not let us film their line. Now what?
Film the FAT at your own facility instead, before the cell ships. You control the site, the customer is already present, and the cell is running the customer's parts. Get the part substitution agreed in advance. If even that is closed off, publish the simulation pane alone with an explicit note that the installed cell is under a confidentiality agreement — that sentence is itself a credibility signal to the next buyer.
Should the cycle time go on screen or in the caption?
On screen, with its conditions, or nowhere. Captions get stripped by reposts and screen recordings, and a number floating without its measurement conditions is the version that ends up in someone's spec. If the conditions are too long to fit legibly, the number is not ready to publish.
Is aerial footage of the plant worth anything at all?
As three seconds of establishing context, yes. As the body of the video, no. The distinguishing information in this category is at the tool centre point, not at 200 feet, and a plant exterior is the one shot a competitor can trivially match.