Developing an eco-driving strategy in a hybrid traffic network using reinforcement learning.

Area of Science:

• Traffic Engineering
• Artificial Intelligence
• Environmental Science

Background:

• Eco-driving is crucial for reducing greenhouse gas emissions and improving public health.
• Integrating autonomous vehicles (AVs) into traffic networks presents challenges in data management and traffic control.
• Existing traffic networks are complex, hindering accurate mathematical modeling for eco-driving strategies.

Purpose of the Study:

• To develop and evaluate an eco-driving strategy for hybrid traffic networks with AVs and human-driven vehicles (HDVs).
• To address challenges in real-time data sharing and decision-making for traffic control agents.
• To optimize fuel efficiency and reduce vehicle delays through coordinated AV and traffic light control.

Main Methods:

• Utilized the Simulation of Urban Mobility (SUMO) simulator for computational analysis of traffic data.
• Employed a model-free reinforcement learning (RL) algorithm, proximal policy optimization, for decision-making.
• Introduced varying percentages of AVs (5%, 10%, 20%) into the traffic flow to assess impact.

Main Results:

• The 10% AV penetration rate demonstrated the fastest convergence to optimal average rewards.
• This strategy led to significant reductions in fuel consumption across the traffic network.
• A notable decrease in total delay for all vehicles (AVs and HDVs) was observed.

Conclusions:

• A hybrid traffic network with a 10% AV penetration rate and RL-coordinated traffic lights offers an effective eco-driving strategy.
• This approach enhances fuel efficiency and minimizes overall traffic delays.
• The study highlights the potential of AI in optimizing sustainable urban mobility.