2022.7-2023.1 **(Research)** Decision Making for Connected-and-Automated Vehicles

Introduction

In the field of Intelligent Transportation Systems (ITS), Connected and Automated Vehicles (CAVs) are viewed as a crucial component. By utilizing information communication and cooperative operation, CAVs have the potential to enhance road safety and improve traffic efficiency. As the transportation industry shifts from human-operated vehicles (HVs) to large- scale automated vehicles (AVs), it is important to study the interactions and cooperation between CAVs and HVs in mixed- traffic environments. Previous research has utilized machine learning technology, specifically deep reinforcement learning (DRL) models, to model mixed autonomy CAVs and decide their behaviors. However, there is a limited number of studies that jointly address the cooperation and interactions between AVs and HVs. This research presents a novel attention-enhanced graphic reinforcement learning approach for safe, efficient, and cooperative lane change decision-making in mixed autonomy CAVs. This approach models interactions with a graph neural network (GNN) and incorporates the graph representation into DRL. The graph structure represents the connectivity between AVs and the attention mechanism in the GNN highlights the intensity of influence between vehicles. The proposed approach is verified through a highway ramp simulation scenario created in SUMO. The results of the experiment show that the proposed approach significantly improves CAV performance in terms of safety, efficiency, and cooperation

The schematic diagram of proposed framework.

Tech Stack

• Theory
  • Machine Learning (GNN, Attention Mechanism)
  • Reinforcement Learning
  • Decision Making for Robotics
  • Intelligent Transportation System
• Programming
  • Python
  • Pytorch
  • SUMO