Doctoral Defense - Mohammad R. Hajidavalloo

About the Event

The Department of Mechanical Engineering

Michigan State University

Ph.D. Dissertation Defense

Tuesday, April 22, 2025 at 2:30 PM EDT

Engineering Building Room 2408

ABSTRACT

DEVELOPMENTS ON COLLABORATIVE SENSING AND DATA-DRIVEN CONTROL FOR INTELLIGENT VEHICLES

By: Mohammad Hajidavalloo

Advisor: Dr. Zhaojian Li

Recently, intelligent vehicle systems have significantly advanced through the integration of collaborative estimation, sensing and data-driven control methods, aiming to enhance vehicle safety, efficiency, and stability. These intelligent vehicles leverage collaborative frameworks that integrate data-driven control methodologies, enabling real-time adaptation and proactive decision-making based on continuous analysis of historical and real-time data. This integration significantly enhances the vehicles' abilities to collectively interpret complex environmental interactions, anticipate potential hazards, and optimize performance, thus achieving superior outcomes compared to individual vehicle systems. However, significant challenges remain, primarily stemming from system complexity, modeling uncertainties, and real-time performance requirements. In this thesis, three innovative methodologies are proposed to address these challenges by advancing collaborative sensing and data-driven control for intelligent vehicles.

First, inspired by mechanical systems, we introduce a novel microscopic traffic model based on a mass-spring-damper-clutch analogy. This model effectively characterizes longitudinal vehicle interactions within traffic flow, capturing drivers' car-following behaviors and reaction time delays, while explicitly addressing the bi-directional impact between leading and following vehicles. Stability analyses were done to give the conditions under which string stability holds. Additionally, an efficient online parameter identification algorithm, leveraging recursive least squares with inverse QR decomposition, is developed and validated using real-world driving data and connected vehicle datasets, enabling accurate real-time estimation of driving-related parameters crucial for predicting vehicle trajectories.

Second, we propose a cloud-assisted collaborative estimation framework that employs Gaussian Process (GP) models to enhance road information discovery. Unlike traditional single-vehicle methods susceptible to measurement errors and model uncertainties, our framework integrates multiple heterogeneous vehicle measurements with cloud-based GP estimations. Each vehicle refines its local estimation using "pseudo-measurements" obtained from cloud-based GP outputs, subsequently updating the cloud model in a recursive and collaborative manner. Comprehensive simulations and hardware-in-the-loop experiments confirm significant improvements in estimation accuracy and robustness.

Third, we present a model-free vehicle rollover prevention strategy using Data-Enabled Predictive Control (DeePC), circumventing explicit system modeling challenges. DeePC utilizes historical input-output data to directly predict vehicle behavior, enabling proactive rollover prevention even under challenging driving scenarios. To enhance computational efficiency, a reduced-dimension DeePC employing singular value decomposition is proposed. Extensive validations through high-fidelity CarSim simulations on diverse vehicle types demonstrate DeePC's superior performance over traditional model-based approaches, maintaining vehicle stability while preserving maneuverability under extreme conditions.

Persons with disabilities have the right to request and receive reasonable accommodation. Please call the Department of Mechanical Engineering at 355-5131 at least one day prior to the seminar; requests received after this date will be met when possible.