Time-Critical Digital Twin Infrastructure: A Data Model and Synchronization Analysis Toward Vulnerable Road Users Safety in Road Intersections
Name: VICTOR MANUEL GARCIA MARTINEZ
Publication date: 16/12/2025
Examining board:
Name |
Role |
|---|---|
| CELSO JOSE MUNARO | Examinador Interno |
| DIVANILSON RODRIGO CAMPELO | Coorientador |
| FLÁVIO GERALDO COELHO ROCHA | Examinador Externo |
| JEAN - MARIE BONNIN | Examinador Externo |
| JULIANO ARAUJO WICKBOLDT | Examinador Externo |
Pages
Summary: The increasing complexity of urban mobility and the growing vulnerability of non-motorized road users have
intensified the need for intelligent, time-critical safety solutions at road intersections. This Thesis investigates whether current Digital Twin (DT) technologies and supporting infrastructures can meet the stringent temporal requirements of safety-critical applications in Intelligent Transportation Systems (ITS), with a specific focus on protecting Vulnerable Road Users (VRUs). To this end, a functional framework based entirely on open-source tools is proposed, comprising a replicable reference architecture, a semantic data model based on discrete risk states, and a methodology for assessing worst-case timing constraints. The
Thesis contributes a domain-specific ontology that formalizes the static and dynamic elements of digital road intersections and introduces a lightweight risk classification model that enables real-time inference within deterministic execution environments. A complementary analytical contribution establishes a worst-case network timing model using deterministic network calculus and evaluates synchronization flows over both traditional and Time-Sensitive Networking (TSN)-enabled infrastructures. Experimental validation was conducted using a hybrid DT implementation based on FIWARE and CARLA, deployed on an edge–cloud
platform enhanced with StarlingX. Results demonstrate that the DT-RI concept is functionally viable; however, current non-deterministic communication and computing infrastructures leave little room for the processing of safety-critical operations. Traditional networks were found to be unable to reliably maintain latency guarantees as the system scales with the number of intersections or sensing density. In contrast, TSN- based configurations consistently preserve ultra-low latency even under increased operational demands. Execution at the edge demonstrated acceptable average responsiveness; however, occasional delays were significant enough to compromise the remaining time available for risk inference, rendering cloud-based
processing approaches unsuitable for safety-critical decision loops. These findings confirm that worst-case latency behavior, rather than average performance, should be used in such critical real-time operations. The Thesis, therefore, concludes that future DT-ITS systems must incorporate deterministic communication mechanisms and execute decision logic at the edge to ensure timely reaction. It further discusses that lightweight predictive methods are more suitable than high-complexity AI models when operating under stringent temporal constraints, reinforcing the principle that meeting decision deadlines is more critical than maximizing predictive sophistication.
