City Research Online

Abstract

The increasing presence of autonomous and semiautonomous vehicles in pedestrian spaces, such as rent-to-ride e-scooters, has raised important safety and operational challenges. This study explores the use of neural networks (NNs) for vehicle navigation and control in such environments, using camera images and GPS data as inputs. Specifically, we examine how varying the size of convolutional neural networks (CNNs) influences both classification accuracy and the practical feasibility of real-time deployment. A set of CNN models, inspired by the AlexNet architecture, were trained on a dataset of vehicle-mounted camera images and corresponding driving actions (e.g., accelerate, brake, turn). The evaluation focused on classification performance using AUROC metrics and observed runtime behavior in a simulated environment. While larger models demonstrated stronger predictive accuracy, only smaller networks were capable of real-time operation due to hardware constraints. These findings highlight the trade-off between network complexity and deployability in pedestrian-focused autonomous systems. Additionally, the study underscores concerns around the over-reliance on GPS data and the limitations of vision-only approaches in unstructured environments. Further work is needed to strengthen decisionmaking robustness and support safe, effective AV deployment in shared pedestrian spaces.

Publication Type:	Conference or Workshop Item (Paper)
Additional Information:	© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Publisher Keywords:	Autonomous Vehicle, Micro-mobility, Convolutional Neural Network, AlexNet, Collision Avoidance, Model Complexity
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology School of Science & Technology > Department of Computer Science
SWORD Depositor:	Symplectic Administrator

Export

Downloads