The main achievements in each pillar during the first two years of URANUS are summarized below.
In Pillar 1, we proposed a novel methodology, based on Bayesian Statistics, to predict traffic patterns in areas and times where direct measurements are not available. In addition, we proposed algorithms that reveal urban traffic demand patterns. To support this, we developed a model-based origin-destination (OD) estimation framework that integrates a path-based traffic model within an optimization-based formulation, effectively reconciling observed link-level data with modelled traffic flows. To address data quality challenges from sensor faults, we developed a fault-tolerant algorithm that detects, isolates, and corrects faulty data during OD estimation. Finally, we developed custom, real-time optimization and machine learning solutions for fine-grained traffic state estimation.
In Pillar 2, we developed several model-based traffic control schemes from an optimization and control-theoretic view. The established approaches tackled the challenge of improving UM performance from multiple angles by developing better and faster solution algorithms, analyzing the stability and optimality properties of the network, optimizing multiple objectives, considering mixed traffic (both human-driven and autonomous vehicles), dealing efficiently with uncertainty, and securing the road networks against adversaries. Moreover, we devised UM control strategies that rely solely on traffic data. The developed strategies explicitly (by selecting when each vehicle should depart and what route to take) or implicitly (through pricing) control traffic demand, aiming to eliminate congestion.
In Pillar 3, we developed novel UAV placement and control methodologies to improve UM performance. Specifically, we designed and solved multi-objective optimization problems that select the locations of sensors to maximize traffic flow coverage and maintain minimum distances between sensor locations for fixed sensor locations (e.g. stationary UAVs). In addition, we proposed an online, uncertainty-driven UAV trajectory planning methodology inspired by statistical learning theory, enabling UAVs to select their future sensing locations in real-time.
In Pillar 4, we established a comprehensive framework for generating and analyzing real-life UM datasets. Towards this direction, our work focused on: (a) establishing a regulatory and operational framework for multi-UAV missions in urban settings, (b) developing Computer Vision algorithms for vehicle detection, re-identification, trajectory mapping, and speed estimation, and (c) building the software tool “UAVTrafficPy” for extracting and visualizing traffic data. Moreover, we established an optimization framework for calibrating simulated traffic models and constructed a calibrated simulation model for part of the Nicosia traffic network. Finally, we developed and released “SUMO-UAV-Py”, an open-source plugin integrated into the SUMO simulation environment to emulate UAV-based traffic sensing.
