26Mar2026
08:00 Master's Defense IC room 85
Topic on
AI-Based Solutions for Connected Vehicles
Student
Carnot Luiz Braun Guimarães Filho
Advisor / Teacher
Allan Mariano de Souza
Brief summary
The consolidation of Intelligent Transportation Systems (ITS) and the transition to the Internet of Vehicles (IoV) generate massive volumes of data, requiring computational solutions capable of transforming raw information into actionable knowledge for urban mobility. This dissertation investigates Artificial Intelligence-based solutions for connected vehicles from three fundamental perspectives: vehicle emissions prediction, distributed learning architectures, and decision support via intelligent agents. Initially, the work evaluates machine learning and deep learning models for estimating carbon dioxide (CO2) emissions using real operational data. The experimental results demonstrate that ensemble methods, specifically XGBoost, outperform neural network architectures (RNN, GRU, LSTM) in stability and predictive accuracy in this domain, offering better adherence to short-term fluctuations. Next, the impact of deployment architectures—Centralized Learning, Federated Learning, and Split Learning—on scalability, latency, and privacy is analyzed, highlighting the trade-offs between communication cost and performance in scenarios with non-independent and identically distributed (non-IID) data. Finally, the study explores the use of agents based on Large Language Models (LLMs) to mitigate the semantic blindness of classical routing algorithms, proposing hybrid architectures that integrate contextual reasoning and deterministic verification. This research contributes to the development of more sustainable and adaptive vehicular systems, aligning operational efficiency with edge constraints and environmental requirements.
Examination Board
Headlines:
| Allan Mariano de Souza | IC / UNICAMP |
| Ademar Takeo Akabane | PUC-Campinas |
| Edmundo Roberto Mauro Madeira | IC / UNICAMP |
Substitutes:
| Islene Calciolari Garcia | IC / UNICAMP |
| Roberto Sadao Yokoyama | CECS / UFABC |