HyperQ AI Safety deploys on existing CCTV in under 1 hour, using ONVIF auto-recognition to connect to installed camera infrastructure. For a robot cell, that 1-hour setup establishes a continuous monitoring layer around the cell envelope — detecting human presence inside the safety perimeter, unexpected objects in the working path, and anomalous conditions during test mode or maintenance access. The hardware controls required by ISO 10218 and ISO/TS 15066 remain in place. The AI monitoring layer answers the question those controls cannot: what has happened in the cell in the 364 days between risk-assessment reviews?
Robot cells get assessed at commissioning. The risk assessment is thorough: the system integrator has followed ISO 10218-1 (robot manufacturer requirements) and ISO 10218-2 (integration requirements), the guarding is designed and validated, the safety-rated monitored stop and protective separation distances are calculated for collaborative or fenced operation, and the documentation file is complete. The cell is commissioned with confidence.
That confidence then ages without verification.
Twelve months later: the same cell, the same robot, a new shift pattern added in February, two staff changes in the cell-operation team, a tooling modification in April that the maintenance team handled without updating the risk assessment, and a test-mode procedure that the newest technician learned from the previous technician informally rather than from the written procedure. The risk assessment in the file still says 2024.
The cell has changed. The record has not.
What the cell envelope means
The safety perimeter around a robot cell is not a static physical boundary — it is a defined zone within which the robot's range of motion and maximum payload create a hazardous area. Under ISO 10218-2, the guarding is designed to prevent access to that zone during automatic operation. The safety-rated stop, the light curtain or hard fence, the interlocked gate — these are the engineering controls that make the cell safe when they function as designed.
The critical word is "when." Engineering controls function as designed under the conditions that existed when they were designed. Cell modifications, tooling changes, new programming that extends the working envelope beyond the original assessment, test-mode procedures that bypass normal safety inputs, and maintenance access patterns that have evolved informally since commissioning all represent departures from the conditions under which the risk assessment was valid.
ISO 10218-2 requires a risk assessment update when the robot system is modified. In practice, minor tooling changes, software adjustments, and operational procedure variations are handled at the maintenance or programming level without triggering a formal reassessment. The gap between the regulatory requirement and the operational reality is not unique to robot cells — it is the same gap that exists in guarding compliance, LOTO management, and chemical handling procedures across manufacturing. The difference with robot cells is that the energy involved makes the consequence of an unexpected entry severe.
Where the monitoring gaps are
Three specific gap periods emerge consistently in robot cell safety incident analysis:
Test mode and fault recovery. When a cell faults and stops during production, the recovery procedure often involves placing the cell in a reduced-speed or test mode that allows manual interaction with the robot while it is partially energised. ISO/TS 15066 Clause 5.4 (speed and separation monitoring) establishes that the minimum protective separation distance between a robot and a person is a function of robot speed, braking distance, and human approach speed — at reduced test speed, the calculated separation distance is shorter but is not zero. For fenced robot cells with test-mode provisions, the written procedure specifies who may enter under what conditions. The practical gap: test-mode access is time-pressured (the line is stopped, production is waiting), the procedure may be 18 months old with no revalidation since the last tooling change, and the person executing the recovery is often the most experienced technician on shift — whose confidence that they can handle it safely is exactly what makes procedure shortcuts likely.
End-of-shift and changeover. Cell loading and unloading at end of run or product changeover involves entry at the payload zone boundary. The procedure is routine and the risk is managed by training and habit, but routine is where informal workarounds develop. A second worker stepping inside the fence to assist with an awkward fixture while the first worker is still at the control pendant, a tooling change done with the power on because the changeover has been done 200 times without incident — these are the boundary-case events that are invisible until they coincide with a failure mode.
Night and weekend maintenance. Maintenance access to robot cells outside normal production hours is a reduced-supervision period. The permit-to-work process applies, but supervision is lighter and the maintenance team has a strong incentive to minimise downtime. LOTO compliance, test-mode protocol, and working-alone provisions are all at higher risk of shortcut during this window.
What AI monitoring adds to the hardware controls
HyperQ AI Safety monitoring around a robot cell works as a verification layer for the three gap periods above. The camera coverage of the cell area detects:
- Human presence inside the guarded perimeter during automatic operation (hard breach: cell should stop or has stopped; monitoring confirms who entered and when)
- Multiple persons inside the cell perimeter when procedure specifies one person
- Person remaining in cell zone after production restart (LOTO clear signed but person still inside)
- Unexpected payload or object in the working path during test mode (AI object detection on the cell interior camera)
- Duration of test-mode access events (baseline: test-mode recovery is expected to take 5–12 minutes; an event lasting 45 minutes in test mode is anomalous)
A Busan-based manufacturer that deployed HyperQ AI Safety on existing CCTV across its production facility — going live in under 1 month — used event detection as a continuous verification layer alongside its hardware safety controls, providing an audit record for every zone-access event across its operational equipment areas.
Robot cell monitoring scope: what the system covers and what it does not
| Monitoring scope | AI monitoring on CCTV covers | Not covered by camera monitoring |
|---|---|---|
| Cell envelope access | Human presence in guarded zone — timestamped, alerted | Interior conditions not visible from external camera |
| Test-mode procedures | Duration and person count of test-mode entries | Correct execution of procedural steps |
| LOTO transition | Person in zone at LOTO sign-off or production restart | Physical lockout device status |
| Changeover access | Number of persons in zone, access duration | Tool condition, fixture seating quality |
| Maintenance window | Person in zone, duration, shift window | Atmospheric conditions, mechanical state |
| Payload anomaly | Unexpected object in working path (camera angle permitting) | Object weight, material properties |
The distinction is operational: AI monitoring on CCTV provides a continuous access record and anomaly detection at the perimeter and interior camera angle. It does not replace proximity-sensing safety hardware (light curtains, safety mats, area scanners) required under ISO 10218-2 or the speed-and-separation monitoring required for collaborative robot applications under ISO/TS 15066. Those controls are the engineering barrier. The AI monitoring layer is the verification and documentation record.
Standards context
ISO 10218-1 and ISO 10218-2 define safety requirements for industrial robots (Part 1) and robot systems and integration (Part 2). ISO 10218-2 Clause 5.4 requires a risk assessment for the complete robot system, and Clause 5.6 requires that the safeguarding be designed to prevent access to the danger zone during automatic operation. Any modification to the system — tooling change, new end-effector, software update that changes speed or reach — requires a review of the applicable risk assessment clauses.
ISO/TS 15066 provides supplementary requirements for collaborative robot applications, including speed and separation monitoring, power and force limiting, and hand-guiding modes. For cells that operate in a collaborative mode alongside workers, the separation distance calculations are based on robot speed, worker approach speed, and stopping time. Any change in robot speed parameters or cell layout changes the separation distance calculation.
Neither standard specifies how the cell envelope should be continuously monitored between formal risk-assessment reviews. The gap between the regulatory requirement (update on modification) and operational practice (minor changes handled at maintenance level without formal reassessment) is where continuous monitoring provides value — not as a substitute for the required documentation, but as an evidence record that the cell's access control was functioning throughout the period.
Singapore — WSH (Machinery) Regulations and WSH Act
The WSH (Machinery) Regulations require that machinery guarding be maintained in effective working condition. The WSH Act places the general duty on the employer to ensure safe systems of work. For robot cells, this includes not only the physical guard condition but the procedural controls (test mode, LOTO, changeover) and evidence that monitoring of those controls is active.
Malaysia — Factories and Machinery Act and OSH Act 1994
The Factories and Machinery Act regulates machinery installation and operation, including guarding requirements. DOSH inspection of robot cells covers both the physical guard condition and the operational safety procedures. A continuous access event record for the cell demonstrates ongoing monitoring of the safety system, not only periodic inspection.
Configuring cell monitoring
Three inputs are required: the camera or cameras covering the cell area, the zone boundary drawn around the cell perimeter, and the production/test-mode state. For cells with a PLC safety relay output that indicates the cell state (auto/test/LOTO), that signal can be integrated directly. For cells without that integration, a manual schedule or state-input button at the cell pendant provides the state input.
Camera placement matters more for robot cells than for most other monitored zones. A single overhead camera covering the full cell interior is the most informative setup. Perimeter cameras (looking across the fence from outside) capture entry events but not interior conditions. Where the brief calls for payload anomaly detection, an interior camera with line of sight to the working envelope is required.
The first week of operation typically requires zone boundary calibration to exclude the robot arm itself from triggering person-detection events (the robot is not a person). This is a zone masking step in the configuration interface and takes less than 30 minutes once the camera angle and robot motion range are established.
Smartband monitoring for robot cell technicians
For the maintenance and test-mode gap periods, wearable monitoring for the technician or operator inside the cell adds a dimension camera monitoring cannot cover: physiological state during the access event. HyperQ AI Safety supports smartbands at $35–250 per unit, logging heart rate, SpO2, and skin temperature.
A technician conducting a test-mode recovery under time pressure (line stopped, production supervisor waiting) shows elevated heart rate in the vitals log for the duration of the access event. If the access event is also a zone breach — entered without completing the written sign-off step — the combined record shows: zone breach at 14:23, heart-rate elevation from 72 to 118 bpm over 8 minutes, normal return after exit. That record is more informative for a post-incident or near-miss review than a camera clip alone. It establishes that the physiological conditions associated with error-prone behaviour were present during the procedural gap.
For organisations that require man-down detection inside an enclosed cell (a worker who loses consciousness inside a fully enclosed robot enclosure is not visible from a perimeter camera), the smartband's fall-detection and inactivity-alert functions provide coverage that camera monitoring cannot.
Where this approach has limits
Camera monitoring of robot cells has the same line-of-sight constraint as all camera-based monitoring. Enclosed cells, cells with opaque guarding, and cells where the working envelope is not visible from available camera angles have limited coverage. Interior camera placement (inside the cell, looking across the working path) requires the camera to survive the operating environment — welding spatter, grinding dust, and solvent vapour are common challenges in the cell environments where monitoring matters most.
HyperQ AI Safety does not provide proximity sensing, speed monitoring, or force-limiting functions. It is not a collaborative robot safety system in the ISO/TS 15066 sense. It does not replace the safety-rated components required by ISO 10218-2. Any plant that replaces required hardware safety controls with camera monitoring is creating a more hazardous cell, not a safer one.
For cells that have never had a formal risk assessment update since commissioning, the continuous monitoring record is useful but does not substitute for the update. If tooling changes, software changes, or operational procedure changes since commissioning have not been captured in a risk-assessment review, the correct action is to conduct that review — the monitoring record provides useful input to the review but does not perform the engineering analysis that the standard requires.
Closing
A robot cell is safe when its hardware controls function as designed and its procedural controls are followed consistently. The hardware controls are verifiable by inspection. The procedural controls are verifiable by continuous monitoring.
The risk-assessment file documents what the cell was designed to be. The continuous access event record documents what the cell has actually been — every test-mode entry, every changeover access, every night-shift maintenance window — across the 364 days between formal reviews.
The most dangerous period for a robot cell is not the commissioning week. It is month 14, when the newest technician has learned the test-mode recovery procedure from a colleague rather than from the document, and the last formal review was before the tooling change in April.
Worker zone monitoring and AI-enforced access control in manufacturing · EV battery lines and the safety assumptions they break · HyperQ AI Safety
Send us your robot cell camera layout and the three highest-risk access scenarios your EHS team has identified — test mode recovery, changeover, or maintenance window — and we'll return a monitoring configuration plan with zone definitions, state-integration options, and a 30-day event record template mapped to your ISO 10218 or WSH audit requirements. No contract until you've reviewed the configuration against your actual cell. Send us your cell layout