Keywords

Connected and automated vehicles, Collaborative Heterogeneous perception, Codebook Compression, Collaborative object detecion

Abstract

Recent advances in computational and communication technologies have enabled deep neural networks (DNNs) and high‐speed vehicular wireless links, paving the way for cooperative perception and cognition in autonomous systems. By exchanging complementary observations, these methods overcome inherent sensor limitations, such as occlusions and restricted range, and thereby improve the detection of partially hidden targets. However, real‐world deployments face challenges from noisy measurements (e.g., GPS inaccuracies) and heterogeneous perception stacks: vehicles often rely heavily on visual detectors, yet there is no standardization ensuring uniform object‐detector architectures across a fleet.

In this dissertation, we first explore the design of a lightweight detector for small and moving objects and then review various map representations used in autonomous driving.

In the main part of this thesis, we introduce a unified framework for collaborative perception that supports both communication efficiency and architectural diversity. Our approach enables vehicles with different sensor configurations and neural network architectures to effectively share and interpret intermediate perception data. To address bandwidth limitations, the framework employs a compact representation of shared features, allowing agents to collaborate without overwhelming communication channels. Furthermore, it is designed to easily accommodate new agents joining the system without requiring extensive retraining. Through extensive experiments, we demonstrate that this approach achieves robust perception performance across a range of scenarios, offering a scalable and practical solution for real-world multi-agent autonomous systems.

Completion Date

2025

Semester

Fall

Committee Chair

Fallah, Yaser

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Computer Science

Format

PDF

Identifier

DP0029721

Document Type

Thesis

Campus Location

Orlando (Main) Campus

Download

Share

COinS