Reinforcement learning-based car tracking model could reduce fuel consumption

The transportation sector remains one of the largest sources of air pollution and climate change on Earth, accounting for approximately 59% of oil consumption and 22% of CO emissions.₂ Identifying effective strategies to limit vehicle fuel consumption could thus help reduce pollution while addressing global energy shortages.

Researchers at the Hong Kong University of Science and Technology recently set out to address this challenge using a computational model based on reinforcement learning.

This model, described in an article published on the preprint server arXivis designed to optimize fuel consumption in car following scenarios, especially in situations where semi-automated and autonomous vehicles are driving in close proximity and need to maintain a safe distance from each other by adjusting their speed.

“The inspiration for this report came from the growing demand for sustainable and energy-efficient transportation solutions,” Hui Zhong, co-author of the report, told Tech Xplore. “With traffic congestion and inefficient driving behaviors contributing significantly to fuel consumption and emissions, we sought to find ways to mitigate these challenges.”

The main goal of Zhong and his colleagues’ recent work was to develop a computer model that would optimize fuel consumption in car-following scenarios, while ensuring that cars maintain a safe distance from each other and that traffic flows efficiently. The model they developed, dubbed EcoFollower, is based on deep reinforcement learning.

“EcoFollower is a reinforcement learning-based car-following model designed to optimize fuel consumption while driving,” Zhong explains. “The model continuously learns from its environment, adjusting following distances and acceleration patterns to achieve the most fuel-efficient driving behavior. What sets EcoFollower apart is its ability to balance fuel efficiency with maintaining smooth and safe traffic flow.”

Conventional models designed to optimize vehicle operation in vehicle following scenarios usually focus only on safety or aim to facilitate traffic flow. The EcoFollower model, on the other hand, is designed to also reduce fuel consumption.

The researchers evaluated their model in a series of tests where they applied it to the Next Generation Simulation (NGSIM) dataset. This is an open-source collection of traffic data collected from four different locations. The results of the team’s initial tests were very promising, as EcoFollower was found to significantly reduce fuel consumption in all the scenarios it was tested on.

“We demonstrated that reinforcement learning can be effectively applied to real-world driving scenarios to reduce fuel consumption,” Zhong said. “Our experiments showed that EcoFollower could reduce fuel consumption by 10.42% compared with real-world driving scenarios. This result has important implications for reducing global emissions and promoting sustainable transportation.”

In the future, the EcoFollower model could be integrated into advanced driver assistance systems (ADAS) and autonomous driving systems, helping to increase their efficiency and reduce their environmental impact. In the meantime, the researchers plan to continue working on the model to further improve its performance.

“Although its performance is already superior to that of traditional intelligent driving mode (IDM) and reduces fuel consumption by 10.42% compared with real driving scenarios, more scenarios and datasets are needed to further test and improve its generalization and robustness,” Zhong added. “For example, in a mixed-autonomy traffic environment, the behavior of human-driven vehicles differs from that of autonomous vehicles, which could impact the model’s performance.”

© 2024 Science X Network