3 Safety Layers You Must Check Before Deploying AMRs

The mobile robot market is no longer paying for autonomy alone, and is placing ever greater value on repeatable safety. For a company planning its first robots, this changes the question: not “how smart is this AMR”, but “is the safety of our AMR deployment built in layers”. In practice, a mature deployment rests on three layers — safety software, human supervision and a physically legible environment. This article shows how to check each of them before you sign the order.

The direction is confirmed by consolidation in the industry. As Modern Materials Handling reports, Fort Robotics has acquired Mapless AI, combining expertise in remote supervision and the functional safety of vehicles. The signal is clear: the advantage is no longer an individual robot, but a complete system in which the machine, the operator and the space work together.

Layer 1: safety in software

The first layer is how the robot behaves in situations that cannot be fully foreseen. An AMR must stop correctly in front of an obstacle, plot a route around an unexpected blockage and respond to exceptions — that is, events outside the standard driving scenario. This is not a question of “does the robot drive”, but of “what does it do when something goes wrong”.

At the supplier selection stage, it is worth breaking this layer down into specific questions. How does the vehicle behave when a person suddenly steps onto the route? How does it react when the scanner detects a partial obstacle, such as a protruding pallet? What happens if connectivity is lost or there is a localisation error? The answers should be repeatable and documented, not dependent on configuration “on site, later”. A well-run pilot tests these scenarios deliberately, introducing controlled obstacles and observing whether the response is the same every time.

The software layer is necessary, but it is not enough. Even the best-designed stopping logic only works if the robot has a legible, orderly environment to interpret. We will come back to this in the third layer.

Layer 2: a human in the supervision loop

The second layer is a model in which there is a human behind the fleet — not next to every vehicle, but overseeing the whole. The key change of recent years is that one remote operator can supervise many vehicles at once and step in only when a robot reports a situation it cannot handle by itself. It is this human-in-the-loop model that underpins the real scaling of fleets: it allows the number of robots to grow without a proportional increase in the number of people watching over them.

For anyone planning a deployment, this raises several organisational questions that are easy to overlook in the excitement over the machine itself. Who takes over control when a robot stops and asks for help? How quickly does that person have to respond so as not to block the flow of material? Does the supervision station have access to images from the shop floor and unambiguous information about which vehicle needs a decision and where? The supervision layer is not an add-on to buying a robot — it is part of the process, and it has to be designed in parallel with the routes.

The supervision model works better the fewer “exceptions” the facility itself generates. If the operator is called every few minutes to untangle situations that better markings could have prevented, the benefit of scaling disappears. Human supervision is meant to handle rare cases, not chronic disorder.

Layer 3: a physically legible environment

The third layer is the one most often underestimated, and it determines whether the first two have anything to work with. The principle is simple: if a person cannot understand the robot’s route in two seconds, the environment is not ready. The robot can “see” a zone in its system, but a worker on the shop floor must also recognise it instantly — otherwise friction arises wherever the machine, pedestrians and trucks meet.

This layer rests on three check questions worth asking during a walk-round of the facility:

• Are the AMR routes separated from pedestrian traffic? A shared aisle without a clear division is the most common source of situations in which the robot has to brake and the supervising operator has to intervene.
• Are the collision zones visually obvious? Route intersections, drop-off points and pick-up points must be recognisable from several metres away, not only once you reach them.
• Do the markings also work in poorer lighting? The night shift, dimly lit areas of the warehouse and sections without windows are precisely the places where legibility most often breaks down.

In practice, this layer is built starting with high-contrast floor markings and hard boundaries where the risk is greatest. We begin deployments by marking out and fencing off zones — with ZonePro mobile barriers, which allow the vehicle’s path to be separated quickly without any intervention in the floor, and with well-organised pedestrian protection at crossings and refuges. Only on a surface prepared in this way do AGV and AMR robots move in a way that people can read at first glance — rather than after a training session that nobody remembers a week later anyway.

AMR deployment safety: how to combine the three layers into one readiness audit

The three layers are not a ranking of importance — they are a checklist in which leaving out one element weakens the whole. A robot with excellent stopping logic will not help if nobody knows who takes control of it after an emergency stop. An efficient supervision model turns into constant firefighting if the facility generates exceptions through illegible markings. And the best-marked space achieves nothing if the vehicle itself does not respond correctly to a person stepping onto the route.

So before you order AMRs, go through the whole thing as one readiness audit rather than three separate topics:

1. Software: ask the supplier for documented, repeatable responses to an obstacle, a person stepping in, loss of connectivity and a localisation error.
2. Supervision: design the human-in-the-loop station — who responds, within what time and with what information about the vehicle.
3. Environment: check route separation, the legibility of collision zones and the visibility of markings in poorer lighting; physically fence off the highest-risk zones.

Repeatable safety does not come from a single, smartest robot. It comes from the machine, the operator and the space all playing by the same, understandable rules. It is the cheapest investment you can make before a deployment — and the most expensive mistake you can make if you skip it.

Which layer should you start with if the budget is limited?

With the third layer, i.e. a legible environment. Route markings, separation of pedestrian traffic and fencing off collision zones are the cheapest, deliver results regardless of the choice of robot supplier and reduce the number of exceptions that later burden the supervision layer. This is the foundation without which the other two layers make no sense at all.

Is the supervision layer needed with a small fleet?

Yes, although on a smaller scale. Even with one or two vehicles, you need to establish unambiguously who takes over control after an emergency stop and how quickly they must respond. The difference is that with a small fleet the role of the supervising operator is usually performed by a designated member of the shop-floor staff rather than a separate station — but the rule must be written down, not assumed.

Sources

• Modern Materials Handling, “Fort Robotics acquires Mapless AI”: mmh.com