CALENDAR

Loading Events
  • This event has passed.

« All Events

Faculty Candidate Research Presentation: Lokesh Das

February 29 @ 11:45 am 12:30 pm

Faculty Candidate Research Presentation
February 29, 2024 11:45 am
In-person in Suite 260 large classroom or Virtual via teams

“A Machine Learning-Driven Approach for Next Generation Traffic Control System for Autonomous Vehicles”

Lokesh Das
Ph.D. candidate
Department of Computer Science, University of Memphis
Advised by Dr. Myounggyu Won

Abstract: Traffic congestion is one of the major problems in the United States of America. It cuts precious time and causes a huge loss in the US economy. Traffic congestion also ignites environmental pollution by releasing greenhouse gases. In addition, human drivers experience prolonged exposure to stress due to traffic congestion which often leads to road rage, driver aggression, and traffic fatalities. Autonomous vehicles (AVs) equipped with vehicle-to-everything (V2X) communication and Artificial Intelligence (AI) technologies are becoming the potential game changer for improving traffic flow and driving safety.

In this presentation, I will introduce our work on an advanced traffic control system for autonomous vehicles by leveraging machine learning techniques to improve traffic efficiency and safety. Our framework consists of an intelligent adaptive cruise control system (ACC) and a cooperative lane-change system. We introduce a novel AI-based ACC system that dynamically adjusts the ACC settings based on real-time traffic conditions. This system significantly improves traffic efficiency over existing static model-based approaches. However, we realize that state-of-the-art intelligent ACC systems primarily focus on traffic flow enhancement overlooking the effect of dynamic adjustment of inter-vehicle gap on driving safety and comfort. We develop a Safety-Aware Intelligent ACC System (SAINT-ACC) which effectively assesses driving safety by dynamically updating safety model parameters with varying traffic conditions. This novel approach prioritizes driving safety and comfort along with traffic efficiency.  Furthermore, we present a novel multi-agent reinforcement learning-based intelligent lane-change system for autonomous vehicles that optimizes both the local and global performance of a roadway segment by incorporating local and global traffic information.  By combining the AI-based ACC system and the MARL-based intelligent lane-change system, our next-generation traffic control system for autonomous vehicles aims to revolutionize traffic management, offering a safer and decarbonized future traffic ecosystem.

School of Applied Computational Sciences

3401 West End Avenue, Suite 260
NASHVILLE, TN 37203 United States