Zillur Rahman, M.S. Candidate

Department of Electrical and Computer Engineering

Implementation of ADAS and Autonomy in UNLV Campus

Committee Members:

• Dr. Brendan Morris, Committee Chair
• Dr. Venkatesan Muthukumar, Committee Member
• Dr. Mei Yang, Committee Member
• Dr. Shaikh Ariffuzzaman, Graduate College Representative

Abstract

• The integration of Advanced Driving Assistance Systems (ADAS) and
  autonomous driving functionalities into contemporary vehicles has notably
  surged, driven by the remarkable progress in artificial intelligence (AI).
  These AI systems, capable of learning from real-world data, now exhibit the
  capability to perceive their surroundings via a suite of sensors, create
  optimal routes from source to destination, and execute vehicle control akin
  to a human driver.
  
  Within the context of this thesis, we undertake a comprehensive exploration
  of three distinct yet interrelated ADAS and Autonomy projects. Our central
  objective is the implementation of autonomous driving(AD) technology at
  UNLV campus, culminating in the introduction of novel enhancements to
  essential ADAS modules. These innovations encompass a longitudinal control
  system, an augmented perception system, and a lane detection model.
  Finally, a full-scale implementation of all modules, specifically tailored
  for the UNLV campus environment.
  
  The first project involved setting up an emergency braking system using 3D
  object detection based on monocular vision. We design a simple longitudinal
  PD controller that considers how close pedestrians are and triggers the
  brakes, which the car's drive-by-wire system uses to control the brakes. We
  also augment autonomous car perception systems by eliminating a blind spot
  created by roadside obstacles like walls utilizing an infrastructure camera
  and a trajectory prediction model.
  
  For our second project, we focus on creating a lane detection and
  classification model that works effectively on the challenging streets of
  Las Vegas. These streets pose particular challenges for existing deep
  learning lane detection models, characterized by suboptimal lane-to-road
  contrast, sparse lane markings, and extreme lighting scenarios. To overcome
  these obstacles, we collect data from intricate scenarios, fine-tune
  several advanced models, and develop a model that works better on these
  unique roads. We also add a lane classification branch that provides lane
  marking types like solid or dashed. Besides, we investigate the effect of
  mixed-precision technique in reducing the inference time.
  
  In our final project, we bring self-driving technology to the UNLV campus.
  We employ the Autoware Universe software stack and adapt it to our specific
  sensor setup and vehicle interface. Our platform relies on HD maps for
  localization and navigation. It uses a 3D LiDAR object detection model for
  pedestrians, vehicles, and other dynamic objects. Finally, it uses a model
  predictive controller to implement the generated motion command. Our
  equipped vehicle can navigate autonomously within the campus while adhering
  to traffic rules.