What is actually in service
Police robotics is older and less exotic than the phrase suggests. Bomb disposal units have been in service for decades, and tactical robots used to enter a building ahead of officers are well established. Both are remote-operated: a human controls the machine continuously, and the robot makes no decisions of its own. Their value is straightforward, which is that they put a machine rather than a person into a space that might kill them.
The newer categories are where the questions start. Quadruped ground robots have been trialled by several forces for reconnaissance in hazardous environments. Semi-autonomous patrol units, more common in the US than the UK, have been deployed in transit hubs, car parks and shopping centres, typically providing camera coverage and an audible presence rather than any enforcement capability.
Aerial systems are the largest category of police robotics by deployment volume, but they are treated separately here because their regulatory position is quite different: see surveillance and smart cities for drones.
Remote operation and autonomy are different problems
These get conflated constantly, and they should not be. They raise almost entirely separate questions.
A remote-operated robot is a tool. A human decides what it does, moment to moment, and remains as responsible for the outcome as they would be holding any other piece of equipment. The questions are about proportionality and use of force, and they are questions policing already has a legal framework for, even where applying it to a machine is novel.
An autonomous system decides something for itself. Even where that decision is only navigational, choosing a route, deciding what to approach, an accountability gap opens that existing law addresses poorly. If a patrol robot's route repeatedly takes it past one set of addresses and not another, someone has effectively made a policing decision, but it may be genuinely unclear who: the developer who wrote the routing logic, the force that set its parameters, or nobody in particular.
In practice, almost all policing robotics today sits firmly on the remote-operated side. It is worth being clear about that rather than describing the field as more autonomous than it is.
Force, and the San Francisco reversal
The question of whether a police robot may apply force has been tested most publicly in San Francisco. In 2022 the city's Board of Supervisors approved a policy that would have permitted police robots to use lethal force in narrowly defined extreme circumstances. Following substantial public opposition, the board reversed the decision within roughly a week and sent the policy back to committee.
The episode is instructive for reasons beyond its outcome. The policy surfaced at all because San Francisco has a Community Control Over Police Surveillance ordinance requiring council approval and public disclosure before police acquire or adopt surveillance technology. Without that requirement, the same policy might have been adopted internally with no public debate whatsoever. This is the clearest available demonstration of why procurement and acquisition oversight matters, discussed further under procurement.
The precedent usually raised in these discussions is a 2016 Dallas case in which police used a bomb disposal robot to deliver an explosive charge, ending a standoff. That was a remote-operated machine used in an improvised way by officers who decided to do it, not an autonomous system, and the distinction matters when the case is cited as evidence of what robots may do.
Where the rules sit
There is no dedicated legal framework for police robotics in the UK. A remote-operated robot is treated, broadly, as equipment: its use falls under existing use-of-force law, human rights obligations where a protected right is engaged, and force policy. Where a robot carries cameras, data protection law applies to what it records in the same way it applies to any other police camera.
Autonomous systems are a genuine gap. The EU AI Act classifies certain law enforcement applications as high risk and imposes obligations accordingly, which is the most developed framework anywhere, though its application to physical autonomous systems in policing specifically is still being worked out. In the US, the position varies by state and city, with CCOPS ordinances providing the most substantive constraint where they exist.
The question that is not technical
A recurring theme in public reaction to police robotics, including the San Francisco case, is that objections are rarely about capability. A quadruped robot doing reconnaissance in a collapsed building attracts almost no opposition. The same machine patrolling a residential street attracts a great deal.
That difference is not really about the technology, which is identical. It is about what the presence of a machine in a public space signals, and whether policing by remote presence changes the relationship between a force and the people it polices in a way that is difficult to measure and easy to dismiss as sentiment. It is worth naming as a real consideration rather than treating public discomfort as a communications problem to be managed.
Follow the coverage
PoliceAI News tracks police robotics as it develops: new deployments, force policy decisions, council approvals and refusals, procurement records and the debates that follow. The feed refreshes every 30 minutes.
View Robotics Stories