Interpretable Context Retrieval for Optimising Urban Traffic (I-C-ROUTE)

Person Photo

Professor Nikolay Mehandjiev

Vice Provost (Research)

"I really look forward to leading the team of experts behind I-C-Route, focusing on the application of AI in traffic management, an area I have worked in since 1990. This builds on my prior research in Applied AI and Human-centric AI in collaboration with Airbus, Atos and other blue-chip companies. Prior projects of mine included focus on AI in traffic control and prediction in collaboration with a number of European municipalities, and I-C-Route will build on some of these results to deliver innovations which can change the way in which we travel around Dubai and UAE. I am proud to be able to contribute to this change in collaboration with RTA."

Traffic control and congestion management are increasingly important for Dubai. Indeed, the growth of the transportation network in Dubai is outpaced only by the growth of population and vehicles on the road. Successful traffic management depends on predicting near-future road congestion, and this can be done through comparing current traffic information with retrieved historical traffic patterns on the same or similar roads and junctions. The challenge is that current algorithms are often hard-coded, so traffic managers lack control and the flexibility to discover similarities based on additional domain-specific attributes such as time of the day, precipitation, low visibility due to sand storms, etc.

This project focuses on addressing this problem and delivering a user-controllable traffic pattern retrieval tool embedded in a traffic control AI/LLM system. The system will use semantic search and Retrieval Augmented Generation (RAG) to identify matching historical scenarios, recommend suitable control strategies, interface with specialised forecasting models, and communicate predictions of journey times.

The project will deliver the following outcomes in collaboration with RTA:

  1. A new approach to interpretable embeddings of traffic data as data points in a low-dimensional domain-specific vector space, allowing focus on specific features;
  2. A user-facing tool allowing users to specify features in natural language;
  3. A prototype semantic search RAG system which retrieves past traffic data, using a visual interface to show predicted congestions and strategies for alleviation of these congestions.

The team of senior investigators links AI research in Dubai with UK/European expertise in graph neural networks and symbolic embeddings, aiming to amplify Dubai resources and create world-class capacity for research in neuro-symbolic AI in Dubai.

The project emphasizes ethical, human-in-the-loop AI design with interpretability, bias detection, and user control for transparency. Its scalable architecture supports Dubai’s Cognitive City vision and can be adapted across other urban domains.

Back