Autonomous mobile robots outside the warehouse

Warehouse robots work because the building is designed around them. The interesting question is what happens when that is not true, and the answer varies sharply by setting.

What changed in navigation

Early industrial vehicles followed buried wires or reflective markers. Any change to the route required changing the infrastructure.

Modern autonomous mobile robots build and localise against a map using lidar and cameras, which means they can be deployed into an existing building and re-tasked in software.

That single change moved the technology from a fixed installation to something a facility can adopt without rebuilding.

Hospitals

Among the most successful non-warehouse deployments, and the reasons are specific.

The tasks are transport: linen, meals, waste, specimens, pharmacy items. High volume, repetitive, and currently performed by staff walking long distances.

The environment is structured — corridors, lifts, defined routes — and the building is generally accessible and flat.

The difficulties are equally specific: lifts must be integrated so the robot can call and ride one, corridors are congested with beds and equipment, and the population includes people who are unsteady, distracted or asleep on a trolley.

Deployments generally succeed where the routes are back-of-house and fail where they cross busy clinical areas.

Hotels and buildings

Delivery robots taking items to rooms are a visible application with genuine adoption.

The task is simple, the route is repetitive, and the value is partly novelty and partly saving staff a long walk at an awkward hour.

Lift integration is again the technical crux, and it requires cooperation from the building systems rather than from the robot.

Factories

Moving material between workstations in facilities that were not designed for automation.

Harder than a warehouse because floors are less clean, layouts change, forklifts operate at speed, and human traffic is unpredictable.

Safety standards for industrial vehicles govern here, requiring detection of a person lying down as well as standing, and stopping distances that scale with speed and load.

The commercial case is strongest where the alternative is a person driving a tug on a fixed circuit.

Pavements and last-mile delivery

The hardest of the wheeled applications, and the one with the most attention.

Pavement environments contain kerbs, uneven surfaces, roadworks, parked vehicles, animals, children and weather. Crossing a road requires judging traffic.

Deployments have concentrated on university campuses and planned neighbourhoods, which are pavement environments with unusually favourable properties: wide paths, low traffic and a co-operative population.

Expanding beyond them has proved difficult, and several operators have scaled back.

Regulation is the other constraint: whether a robot is a pedestrian, a vehicle or something else is decided locally and inconsistently.

The problems that recur everywhere

Lifts and doors. Any multi-floor deployment requires integration with building systems, which is a commercial negotiation as much as an engineering one.

Recovery. A robot that becomes stuck needs someone to attend. Remote assistance — an operator who takes over when the machine is confused — is standard practice and is a real ongoing cost.

The honest metric for any deployment is interventions per hundred hours, and it is rarely published.

People. The variable no simulation captures.

Some people ignore the robot, some deliberately block it, some interact with it at length, and children treat it as a toy. Deployments have reported vandalism and theft.

Battery and docking. Continuous operation requires autonomous charging, which requires reliable docking, which fails occasionally in ways that strand the machine.

Fleet management as the real product

An observation that applies across every setting above.

A single robot is a demonstration. A deployment is a fleet, and the software that allocates tasks, schedules charging, monitors health, handles exceptions and reports to the operator is where most of the engineering sits.

Facilities buying autonomy are, in practice, buying that management layer. The vehicle is the commodity.

Which is why integration with existing systems — building management, task scheduling, inventory — determines success more reliably than the navigation stack does.

The pattern that holds

Success correlates with how much the environment can be adapted and how tolerant it is of a stoppage.

A hospital corridor at three in the morning is forgiving. A pavement outside a school at four in the afternoon is not.

Which is the same conclusion the warehouse case produced, arrived at from the other direction.