Cobot proximity to the operator in small appliance assembly — human cadence without stopping the cobot.

The problem

Safe does not mean stopped. And stopped costs money.

The collaborative assembly cell for small appliances — coffee machine, iron, blender, vacuum cleaner — lives with the same trap as almost every cobot installation on the market:

1. Safety is well solved and badly exploited. The certified laser scanner triggers the cobot stop when the operator enters the zone; that complies with EN ISO 10218 and ISO/TS 15066. But between “safe zone at normal cadence” and “danger zone with a stop”, there is a whole middle band where the cobot should slow down, not halt. Almost nobody modulates that band properly.
2. Operator and cobot dance all shift long. Every time the operator’s hand comes closer to place the next part or clear a bit of debris, the cobot stops — because the scanner only knows stop or don’t stop. Human cadence is discontinuous; so is the cobot’s, but for the wrong reason.
3. The missing knowledge is continuous distance, not yes/no. The operator-to-cobot distance is there, physically, all the time. What is missing is a system that measures it continuously, cross-references it with the cycle and turns it into a graded speed command.

The production manager knows this and can do little with classic resources: the scanner is what it is. The health and safety manager does not want to lower their guard. And the cell ends up an expensive machine performing below its potential because the whole system is designed in black and white.

How it fits the IRIS system

iLEAN does not replace the scanner — it adds the layer missing between safe and stopped.

The cobot cadence problem is not a lack of safety: it is a lack of an intermediate layer. The functional safety layer (laser scanner, emergency stop, certified bus) is the one already there, and it is not touched. iLEAN Edge sits on top of it, like the filler between cobot and operator, and contributes a continuous distance the classic system cannot see.

Edge sees the operator in 3D and measures the distance to the cobot continuously. The agent cross-references it with the cobot cycle and triggers the reduced-speed output — before the scanner has to stop anything. The health and safety manager stays in command.

The three iLEAN pieces applied to the collaborative cell:

• Edge — a terminal with a 3D camera and a pose-estimation neural network above the cell. It computes the minimum body-to-cobot distance in milliseconds and translates it into three regimes — normal, reduced, stop — via dry contact or OPC UA to the controller. No facial recognition. The data is per workstation, not per person. Works with no network.
• Connect — captures the SKU in production and the cobot cycle recipe from the PLC or the MES, and absorbs order changes (a new SKU from the importer, a BOM modification) without anyone having to re-send anything.
• Agent — cross-references distance, time, cycle and SKU, proposes adjustments to the shift supervisor (rotating the operator if the workstation goes into reduced regime too often) and generates the per-workstation, per-shift dossier for the risk assessment review. The person validates; the cobot does not reconfigure itself.

See the full IRIS architecture →

Before and after

Binary cobot vs. cobot with iLEAN graded proximity

Aspect	Cobot + laser scanner + binary stop	With iLEAN Edge + Connect + Agent
Operator-to-cobot distance	Yes/no (safe zone or critical zone)	Continuous, in milliseconds, skeleton vs envelope
Cobot response	Normal cadence or stop	Three regimes: normal · reduced · stop
Stops per shift	One for every operator entry	Only the ones that genuinely require it
Facial recognition	n/a	No — analysis per workstation, not per person
Dossier for the risk assessment	Manual reconstruction	Proximity history per workstation and shift
Certified functional safety	Scanner + certified bus	Untouched — Edge sits on top of it

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data from your cell. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Collaborative small appliance assembly cell — coffee machine, iron, blender or similar — with a UR, Doosan, ABB or Fanuc CR type cobot, already in production with a certified laser scanner.
• Edge pilot on one Pareto workstation (3D camera + integration with the cobot controller + dossier). First value expected within a few weeks.
• Expected reduction of unnecessary cobot stops of ≥ 30% against the baseline, with functional safety untouched.
• Indicative payback between 4 and 9 months, depending on the cadence currently lost and the SKU mix.
• Hard lever: cadence recovered hour by hour + a proximity dossier useful for the risk assessment review and for the insurer.

And the health and safety manager’s reasonable doubt

“What if the AI grades the speed wrong and leaves the operator exposed?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI does nothing but measure a continuous distance and translate it into three levels, the best models brought the error below 1.5% ^[1]. And even so, nothing critical is decided alone: the certified functional safety layer (laser scanner, safety bus, emergency stop) stays in place and triggers the stop by itself if Edge fails. The three rings exist for precisely this.

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.