The city of Malaga has become a real urban laboratory to test police routes designed with artificial intelligenceA team from the University of Malaga (UMA) has created a system capable of plotting more efficient patrol routes based on real crime data, with the aim of reinforcing surveillance precisely where it is most needed.
Far from being a simple theoretical tool, this model is based on a detailed virtual map of the city of Malaga This allows for the simulation of patrol strategies before deploying them on the streets. The proposal aims at a very specific objective: to help the police better distribute their resources, avoid neglected areas, and reduce routine routes that are predictable for criminals.
A digital twin for planning police routes with AI
At the heart of the project is the creation of a digital twin of crimes in MalagaThat is, a virtual replica of the city that integrates geographic, social, and security data. This simulated environment captures both the location of criminal incidents and other factors of the urban environment, so that investigators can "test" different patrol configurations without moving a single officer from the station.
According to the work published in the scientific journal Engineering Applications of Artificial Intelligence Titled "Cooperative patrol routing: Optimizing urban crime surveillance through multi-agent reinforcement learning," this digital twin serves to evaluate in advance which surveillance strategy yields the best results. In this way, law enforcement agencies can rehearse multiple virtual scenarios before making operational decisions in the real world.
One of the key contributions of the model is a new indicator, named as coverage ratioThis indicator measures the extent to which patrols are able to protect each area according to its risk level. In contrast to more traditional approaches, which focus primarily on the frequency of patrols or how long an area remains unvisited, this indicator centers on the actual effectiveness of the deployment: how much critical area is actually monitored and with what intensity.
Thanks to this analysis, the tool is able to determine both the ideal number of foot patrols and their recommended routestailored to the crime patterns of each neighborhood. This reduces both "blind spots" without police presence and the waste of resources in lower-risk areas.
50x50 meter grids and patrols treated as intelligent agents
To achieve that level of detail, the research team has divided the city into small grids of 50 by 50 metersEach of these blocks functions as a unit of analysis where the concentration of crimes and other relevant parameters are calculated. This micro-resolution allows for fairly precise identification of which corners, streets, or sections have the highest concentration of incidents.
On this gridded urban board, artificial intelligence simulates the behavior of the patrols, which in the model are represented as "agents" capable of learning and coordinatingThese virtual agents do not follow rigid orders, but rather adjust their movements based on the information they receive from the simulated environment.
The system is based on techniques of multi-agent reinforcement learningIn practice, this means that the AI āātests different routes and strategies, evaluates which approach provides the best coverage, and uses that feedback to refine the routes. It doesn't work with fixed routes, but rather with continuous improvement, where several patrols learn to divide the territory cooperatively.
According to the authors of the study, this approach fits with various criminological theories that argue that unpredictable routes They can discourage criminal activity. If patrols don't always follow the same route or schedule, it becomes more difficult for potential offenders to anticipate where and when police will be present.
AI-powered police routes: fewer repetitive patterns and more coverage
One clear difference compared to more traditional methods is that the system It does not generate a single standard itinerary.Instead, each patrol uses different routes. This variety of routes makes surveillance less monotonous and ensures that high-risk areas are covered from different angles and at different times of day.
Tests conducted in a virtual environment show that, with this model, It increases coverage of so-called critical points.While adjusting the number of officers needed in each area, this avoids both insufficient surveillance in problem areas and the over-deployment of officers in places where it is not as necessary.
The system has been validated in three urban areas of Malaga with different characteristics in size, density, and security level. In all cases, the model was able to adapt to the context and offer more efficient customized patrol routes than the reference approaches with which it was compared.
The researchers emphasize that the goal is not to replace police judgment, but to provide a tool to support decision-makingBased on the simulation results, commanders can assess different scenarios, check how many agents would be needed, and decide which configuration best fits the operational reality of the force.
Although the proposal is still in the experimental phase, the system's behavior suggests that this type of AI-designed police routes This could help to make better use of available resources, especially in cities with high urban density and highly changing security demands throughout the day.
From simulation to the street: real conditions and validation in Malaga
One of the project's strengths is that, despite still being in a In the simulated environment, real operational constraints have already been included.The model includes, for example, eight-hour shifts for patrols and foot patrols, so the suggested routes are not just lines on a map, but paths that a patrol could cover during its workday.
In practice, the tool indicates How many patrols should be deployed in each grid square? and how to distribute their routes so that the coverage index reaches appropriate values āāaccording to the crime risk. This opens the door to adjusting the service planning, shift by shift, based on the accumulated data.
The system has also been tested with Real crime statistics for the city of MalagaBy feeding the digital twin with historical crime records, researchers have been able to test whether the routes proposed by AI improve surveillance of the most problematic areas compared to standard strategies.
The results suggest that technology contributes both to Strengthen police presence in conflict areas such as reducing unnecessary overlap between patrols. In a scenario of limited resources, this fine-tuning is especially relevant to getting the most out of each deployed officer.
The collaboration with the Territorial Intelligence Unit of the Provincial Police Station of Malaga of the National Police This has been key to adapting the model to the reality of police work in the city. This direct contact with daily practice ensures that the system doesn't remain merely an academic exercise, but rather evolves with a view to future application.
Andalusian research and European funding for AI-powered police routes
The development of this system is part of the Juan Palma-Borda's doctoral thesis and is part of the ATREIDES project of the National R&D Plan. The work has been carried out by the Research and Applications in Artificial Intelligence Group of the UMA, a team with previous experience in agent-based models and complex data analysis.
The initiative has included Partial funding from the Ministry of University, Research and Innovationas well as from the European Regional Development Fund (ERDF), through the project "Agent-Based Crime Prediction Models and Data Science." This institutional support positions Malaga as one of the most advanced locations in Spain in the use of AI applied to public safety.
The Discover Foundation has publicized the scope of this research, in which the researcher Eduardo Guzman has highlighted the potential of artificial intelligence to design support tools in police planningThe idea is that law enforcement agencies have objective tools to help them justify and optimize their deployment decisions.
Beyond the city of Malaga, the proposal opens the door to... other Andalusian or European citiesWith sufficient georeferenced information, similar solutions can be adopted. Since it is a grid- and agent-based model, the methodology could be exported, with the necessary adaptations, to other urban environments.
In a context where security and the efficient use of public resources are under scrutiny, this type of project reinforces the commitment to a more data-driven policing management and even less so in intuition or the inertia of historical routines.
Next steps: emergencies, real-time, and more variables in the digital twin
The team from the University of Malaga hasn't stopped at designing patrol routes. The researchers are already working on New lines to expand the scope of the digital twin, with the aim of enabling the system to handle much more dynamic situations.
Among the planned next steps is the development of a module focused on real-time emergency managementThe idea is that, in the event of a specific incident, the system will be able to recalculate recommended routes on the fly, redirect patrols, and propose a reorganization of the deployment to avoid leaving other sensitive areas unprotected.
Also on the table is the inclusion of socioeconomic, environmental and urban mobility factors to the digital twin. Aspects such as the flow of people at certain times, the presence of leisure areas, lighting, trafficāfor example, regulated by the municipal ordinance on low emission zonesā or even weather conditions could influence the calculation of optimal routes.
With more layers of information integrated, the model would become even more realistic and tailored to the daily behavior of the cityThis would allow for better anticipation of where certain types of incidents might increase and for adapting surveillance to urban rhythms that are not static, such as major events, tourist seasons, or changes in mobility.
All this work is part of a broader effort to applying artificial intelligence to citizen security without losing sight of the need for human supervision and compliance with European legal frameworks on data protection and fundamental rights.
With the combination of a detailed digital twin, a coverage index that measures the actual effectiveness of surveillance, and police routes configured using multi-agent reinforcement learning, Malaga is at the forefront of the AI-powered police routesThe project demonstrates that, with quality data and collaboration between universities and security forces, it is possible to redesign urban patrolling to be less predictable, more risk-adjusted, and better aligned with the real needs of citizens.