Amsterdam labs write rules for public robots
Engineers and city lawyers are testing delivery machines with slower speeds, clearer signals, and public appeal channels.
3 min read

The delivery robot that stalled on the Prinsengracht last November did not hurt anyone. It sat across two thirds of a narrow pavement for fifty minutes, until a resident pushed it aside and a technician arrived to reset it. No damage, no injury, no report filed anywhere — and that, more than the machine itself, is what set the current rulemaking in motion.
Amsterdam now has more than four hundred permitted robots operating on public ground: pavement couriers, inspection units in the metro tunnels, mowers in the larger parks. Each was licensed by a different department against a different template. None of those templates agreed on who to call when something goes wrong.
Starting from the incident, not the machine
The working group’s first decision was to stop writing rules about robots. Capability changes every eighteen months; the awkward moments do not. A machine that blocks a route, that fails to yield, that records a face it had no reason to record, that stops dead in traffic — those four situations have been the whole of the complaint log for three years running, across every model on the street.
So the draft regulation is organised around them. For each, it names a maximum response time, a party responsible for arriving, and a log entry that has to exist afterwards. What the machine is, and who built it, is a schedule at the back.
Regulate the failure, not the hardware. The hardware will be replaced before the ink dries.
That framing survived contact with the operators more easily than the group expected. The larger fleets already run internal incident taxonomies that look much like it, because their insurers demanded them first.
The registry problem
Where agreement broke down was visibility. Residents wanted a public register: every permitted unit, its operator, its route, and a number to call. Operators argued that route data is commercially sensitive and that publishing it invites interference — a real risk, since two units were pushed into a canal in the first year of trials.
The compromise splits the register. Identity, operator, and a contact route are public and printed on the machine itself in characters large enough to read from standing height. Route and telemetry go to the city under a data-sharing agreement, available to researchers on request and to the public in aggregate.
It is a smaller win than the campaigners wanted. It is also the first time anyone in the Netherlands can look at a machine on a pavement and find out, in under a minute, who is answerable for it.
What the operators had to concede
Three obligations survived the consultation largely intact, and they are the ones the industry lobbied hardest against.
The first is a yield rule with teeth: on any pavement narrower than two and a half metres, the machine gives way and, if it cannot, it stops and waits. The second is a hard cap on retained video — seventy-two hours unless an incident is open. The third is the one that will cost most, an obligation to fund removal within thirty minutes of a blocking report, which in practice means staffing a recovery crew rather than routing one from a depot.
Operators asked for the removal window to be four hours. The city’s own street teams pointed out that they clear an illegally parked scooter faster than that, and the argument ended there.
What to watch
The regulation goes to the council in the spring. Watch whether the seventy-two-hour retention cap survives its first serious police request, whether the thirty-minute window is measured from the report or from dispatch, and whether the public register lists units that have been withdrawn — because a register that only shows what is currently permitted tells you nothing about who keeps failing.



