From childhood hobby to adult project
Around the age of ten, Rangelov became fascinated by robotics. A book on electrical engineering inspired him to build simple circuits at home using batteries and light bulbs. This early hobby now forms the basis of his PhD research at the University of Twente, where, under the supervision of Maurice van Keulen, he is working on so-called “crime bots.” The project is a collaboration between the University of Twente, the Police Academy, Saxion University of Applied Sciences and the Technical University of Sofia. The strength of this multidisciplinary cooperation comes together in a forensic investigation system made up of several innovative components.
Detecting invisible dangers
At a crime scene, dangers are not always visible. Toxic gases and other hazardous substances may be present without being detectable by smell or touch. To protect officers, Rangelov’s forensic system incorporates advanced gas detection. This uses sensors and AI-powered pattern recognition to quickly identify dangerous substances.
The system can be mounted on a robot, which then enters a high-risk area first. This prevents unnecessary danger. Rangelov refers to a Dutch case in which things went wrong: “Officers stormed a drug lab and fell ill. They had to be taken to the hospital. A gas detection system would prevent such dangerous situations.”
Gas detection not only contributes to safety but also enhances evidence collection. In addition to samples from solid objects, such as walls or doors, airborne evidence can now also be captured. “This enriches the body of evidence as we currently know it,” says Rangelov.
The crime scene in 3D
Capturing a crime scene exactly as it appears at that moment – and being able to revisit it at any time – is what the system’s 3D reconstruction makes possible. For example, experts abroad can remotely view the scene and provide advice.
The technology uses Neural Radiance Fields (NeRF) and 3D Gaussian Splatting – intelligent software that creates a highly accurate 3D model from ordinary photographs. AI enhances the images and ensures that objects are automatically recognised, labelled and stored. This enables the police to quickly compile a complete digital catalogue of evidence.
Because the system is largely software- and AI-based, it is cheaper than the expensive 3D scanners often used today. “The quality is comparable, or even better,” says Rangelov. “And as we continue to improve the algorithms, the system becomes ever more precise.” The technology also has a place in education. Together with colleagues, Rangelov develops educational programmes based on real criminal cases captured in 3D. “Students put on a VR headset, step into the digital crime scene, examine objects and experience realistic scenarios,” he explains. “This allows them to find out during their studies whether the forensic field suits them.”
AI scenarios against tunnel vision
The final component of the forensic system focuses on generating alternative scenarios for what may have happened at the crime scene. Using Natural Language Processing (NLP) and large language models (LLMs), the technology produces multiple realistic scenarios of the possible course of events.
According to Rangelov, this helps to prevent tunnel vision – a phenomenon common in police work that can lead to incorrect conclusions. “Experienced officers often compare a case with earlier ones, but no two cases are ever exactly the same,” he explains. “If you only have that one familiar situation in mind, you risk drawing the wrong conclusion.”
By presenting several plausible scenarios, the system encourages officers to view the evidence more broadly and consider less obvious hypotheses. This increases the chances that crucial clues will not be overlooked.
The future of AI-driven forensic investigation
Rangelov has one major dream for the future: that the forensic system will not only be adopted but will remain in long-term use. He compares this ambition to the development of DNA testing. “At first, DNA tests were experimental and only after years became fully accepted in the forensic world. Now they’re indispensable,” he says. “Hopefully, the same will be true for 3D reconstructions, which could be used by everyone – from car accident investigators to detectives and even archaeologists. The possibilities are endless.”
