Please join the Computer Science Department for the upcoming thesis defense:
Presenter: Pronab Ghosh
Thesis title: Intelligent Vehicle-to-Vehicle Communications with Importance of Fairness and Information Freshness
Abstract: Intelligent Transportation Systems (ITS) showcase cutting-edge services designed to revolutionize transportation and mobility, especially within future smart cities. These services play a pivotal role in bolstering traffic safety, traffic flow management, infotainment, and the dependability of edge-assisted autonomous driving. Consequently, ITS introduces the Vehicle-to-Vehicle (V2V) communication paradigm, facilitating continuous connectivity between moving vehicles and their surroundings. Real-time data exchange regarding acceleration, position, speed, and braking status enables collision avoidance and congestion mitigation. V2V communication streamlines communication pathways, resulting in safer and more comfortable driving experiences, particularly in high-risk scenarios. This thesis investigates two distinct challenges within V2V communications:
1. Multi-Group V2V Communications: This study addresses the establishment and scheduling of data streams and packets between vehicles within a multi-group communication setup. In scenarios involving police cars, ambulances, buses, or city fleets, each group of vehicles communicates within itself. The objective is to establish communication links between all vehicle pairs within a group, utilizing WiFi technology to alleviate the load on cellular networks. Since not all pairs have direct communication capabilities, the problem extends to relaying and scheduling data packets through multi-hop transmissions. Resource blocks, including designated channels and time slots, are allocated. The study aims to maximize communication efficiency among vehicle groups while ensuring fairness and allowing resource block reuse under the SINR constraint.
2. Age of Information (AoI) Minimization: Traditional metrics like throughput and latency do not sufficiently capture data stream timeliness and freshness, critical for autonomous driving and accident prevention. This study targets the minimization of AoI across all data streams in autonomous vehicular networks. The goal is to reduce the total or average AoI over a specified timeframe. Unlike the first study, direct data stream connections between vehicle pairs are absent. Instead, a vehicle broadcasts data to nearby vehicles based on data importance. Minimizing AoI requires optimizing relaying decisions, transmission timing, and data packet dropping. Complexity arises from optimizing nodes for data relaying, transmission timing, and prioritizing newer data packets.
In both studies, mathematical formulations employing mixed-integer linear programming (MILP) are initially employed for optimal solutions. Due to optimization model complexity, scalable heuristic methods are proposed for larger networks. To capture dynamic environmental dynamics, both problems are modeled as Markov Decision Processes (MDP) and tackled using reinforcement learning (RL) techniques such as Q-learning and Double Deep Q-Networks (DDQN). Additionally, hybrid heuristic-based RL methods are introduced to enhance learning behavior and overall performance. Numerical results underscore the efficacy of hybrid approaches in comparison to optimal solutions, random agents, proposed heuristics, and conventional RL methods across networks of varying sizes. In conclusion, this thesis contributes to intelligent transportation systems and future smart cities by offering innovative solutions for vehicular communications. These approaches hold the potential to enhance data transmission efficiency and reliability for autonomous vehicles, paving the way for safer and more responsive autonomous driving experiences.
Dr. Thiago E Alves de Oliveira (co-supervisor, committee chair), Dr. Dariush Ebrahimi (supervisor), Dr. Xing Tan, Dr. Salama Ikki (Electrical Engineering).
Please contact email@example.com for the Zoom link. Everyone is welcome.