Building Autonomous Vehicles for a Sustainable Future

Abstract

This 28-page research investigation evaluates sensory architectures and real-time inference frameworks for small-scale autonomous ground vehicles operating under high-temperature urban road conditions in Qatar. The study analyzes the trade-offs between LiDAR-based spatial point cloud processing versus edge camera-based deep convolutional neural network object detection on power-constrained embedded systems.

Research Problem & Context

Urban transit initiatives in arid climates face specific environmental challenges, including severe thermal constraints on onboard computers and sensor degradation from airborne dust. Standard autonomous vehicle computational pipelines require optimization to operate reliably on resource-limited edge hardware.

Research Question

Can edge-computed computer vision models provide sufficient spatial awareness for low-speed autonomous navigation under high ambient light and thermal constraints compared to low-cost solid-state LiDAR?

Experimental Methodology

Constructed an experimental 1:10 scale autonomous vehicle testbed driven by an NVIDIA Jetson Nano. Evaluated frame rates, memory footprints, and inference latency across YOLOv8 edge vision models versus 2D LiDAR distance scanning across structured test tracks.

Hardware Architecture

• NVIDIA Jetson Nano Developer Kit
• Wide-angle CSI Camera Module
• RP-LiDAR A1 Solid-State 2D Laser Scanner
• Dual H-Bridge Motor Driver & Custom Chassis
• Dedicated LiPo Power Subsystem with Step-Down Regulation

Software & Toolchain Stack

• Python 3.10
• OpenCV 4.8
• YOLOv8 Nano (TensorRT optimized)
• ROS (Robot Operating System) Melodic / Noetic
• Custom C++ PID Motor Controller

Experimental Setup & Testing

Tested on an indoor obstacle course with variable lighting and obstacle placement. Evaluated frame processing and obstacle clearance behavior.

Results & Analysis

Demonstrated that TensorRT-quantized YOLOv8n ran real-time object detection at interactive frame rates on the Jetson Nano, maintaining reliable obstacle detection alongside 2D LiDAR during indoor testing.

Technical Limitations

Experiments were conducted on a 1:10 scale indoor test track; the system did not undergo full outdoor desert field trials, and did not incorporate RTK GPS or multi-lane highway velocity dynamics.

Primary Resources & Manuscripts

• Official Research Manuscript (SharePoint Document) (paper)
• Author Research Portal (external)

Citation Information

@techreport{alagha2024autonomous,
  author = {Alagha, Abedalraheem},
  title = {Building Autonomous Vehicles for a Sustainable Future},
  institution = {Jassim Bin Hamad Technology Stream},
  year = {2024},
  url = {https://abed.page/research/autonomous-vehicles-qatar}
}