Zhiwei Yang
University of Queensland
Supervisors: Associate Professor Zuduo Zheng
Project: Co-operative eco-driving system for mixed traffic on urban roads
About Zhiwei’s research
Zhiwei’s project is about providing vehicle speed advisory to reduce energy consumption or emissions. The technology is termed eco-driving.
How far along is Zhiwei? What has she achieved so far?
Zhiwei is two years along in her PhD. She has proposed and evaluated eco-driving methods based on rules, optimisation, and reinforcement learning (RL) by simulation and real-world trajectory dataset.
Results show that one RL-based method has good comprehensive performance, and temporal and spatial generalsation capability.
What questions will this research answer?
This work will look to optimise all aspects of driving performance, i.e., safety, time efficiency, fuel saving and ride comfort in mixed traffic consisting of connected and autonomous vehicles (CAVs) and human-driven vehicles (HVs).
It is one of the first to propose eco-driving methods based on the most popular and effective RL algorithms and compare them to real-world trajectories and analytical car-following models. It proves that RL is the most promising method to apply in realistic mixed traffic.
What obstacles did Zhiwei overcome? How did she do it?
- Zhiwei needed to run simulations using a newly-developed open-source simulation platform developed in Ubuntu. However without knowlegwhereas of the platform, and without online tutorils, she contacted the developer for assistance. With that help she was able to explore the platform and successfully run her simulations.
- The proposed rule- and optimisation-based methods were not able to deal with uncertainty. Referring to literature, she decided to use reinforcement learning (RL) algorithms.
- Programming to implement the RL algorithms was a challenge, so Zhiwei took online courses to acquire basic knowledge. Then she recycled reliable codes and modified according to her needs.
- The performance of the proposed RL-based models unsatisfactory, so she designed the reward function and tuned the hyperparameters carefully. Moreover, she developed a hybrid method combining the RL algorithms with rules.
What does Zhiwei see as her next move upon completing her PhD?
She would like to find a postdoctoral researcher position to continue doing research, because she I feels she needs more than a 4-year PhD study.
A poster for Zhiwei’s PhD project, made for display at the 2022 ITS Australia awards.
A word from PhD supervisor, Associate Professor Zuduo Zheng
Why is this PhD important to investigate?
Zhiwei’s PhD focuses on eco-driving. In Australia, in which road transport is responsible for about 16% of total greenhouse gas emissions. Some studies show that on current trends, Australia will only reach an emission reduction target of 7 per cent on the 2005 level by 2030, 19 percentage points short of the minimum target mandated by the Paris agreement. Eco-driving, combined with the power of connected and automated vehicles, can be a promising approach in reducing road transport’s impact on energy consumption and emissions.
What are the major challenges to overcome in the field?
There are many challenges in developing eco-driving strategies. Three of them are:
- on top of energy consumption and emissions, other important factors related to driving should also be considered, such as safety, efficiency and comfort;
- the robustness of the eco-driving system in coping with changing traffic conditions, and unexpected disturbances in particular; and
- the effectiveness and even feasibility of the eco-driving system in low penetration rate of connected and automated vehicles.
Where might this work lead in the (near and far) future?
Zhiwei’s work might lead to the real-time implementation of eco-driving strategies as the penetration rate of connected and automated vehicles increases; and her work might also lead to the integration of eco-driving strategies with signal timing of intersections, which might ultimately turn the green light to the green wave for each driver.
Contact Zhiwei
If you’d like to contact Zhiwei about her research, please contact her via her LinkedIn profile.