showSidebars ==
showTitleBreadcrumbs == 1
node.field_disable_title_breadcrumbs.value ==

Improving the safety and reliability of self-driving cars

SMU Assistant Professor Xie Xiaofei aims to help develop Singapore’s Smart City capabilities.

By Stuart Pallister

SMU Office of Research – Autonomous driving systems (ADSs) are complex as they consist of modules such as perception, localisation, prediction, motion planning and control. Each module performs specific tasks which can enable self-driving cars to operate safely and efficiently.

For Xie Xiaofei, Assistant Professor of Computer Science at Singapore Management University, the perception module is of paramount importance as it effectively serves as the ‘eyes’ of the ADS as it allows the self-driving vehicle to perceive and understand its surroundings. 

In their grant application proposal, Professor Xie and his collaborator, Dr Liu Yang of Nanyang Technological University (NTU), state that the perception module serves as a ‘vital link between the vehicle and its environment.’

The research project, funded by a Ministry of Education Academic Research Funding (AcRF) Tier 2 grant, is due to start in August 2024 and is expected to last three years. It aims to assess the reliability and robustness of the perception module, which relies on various sensors including cameras, radar, and light detection (LiDAR) sensors to interpret road and traffic conditions.  

The objective of the project, the grant proposal states, will be to develop new technologies that ‘assess the quality and reliability of the perception module in an ADS with respect to vehicle motion and understand the impact of perception errors on other modules of ADSs such as decision-making.’ 

“Like human beings, self-driving vehicles need to understand the road conditions, the traffic, whether there are other vehicles or obstacles,” Professor Xie told the Office of Research. “So this is the first stage and now the driving system has some basic understanding of the environment. Then you have the planning module. Based on the traffic situation, I need to plan a route to get to my destination. And finally comes the control module, turning left or right based on the perception and plan.”

Understanding ADS

However, software and module issues can have an impact on the robustness of the overall system. Professor Xie points out that, while most studies have focused on the robustness of the perception module, these often overlook the broader impact of perception errors on the entire ADS. 

“So, in this project we will test the perception module but at the same time we will also consider the other modules like planning and control.

“You can make some errors with the perception module but in planning we can mitigate them. However, there are some perception errors that have a significant influence on planning and on the whole system, so we need to understand the relationships and influence of the different modules. That’s our focus.”

According to the grant proposal, the researchers aim to develop advanced error prediction methods to ‘enable proactive mitigation strategies … and enhance the quality of reliability of perception modules in ADSs.’ 

“This is complex. Our focus is the perception module as this is very important, but we will also consider the influence of this module on the others. This is a key difference between our project and other existing projects.” 

The project is expected to yield a series of top-tier journal and conference papers but Professor Xie, whose research has previously focused on software quality assurance, hopes they will also be able to “develop a software system to automatically test self-driving systems.”

Driving the Smart City

He hopes this project ‘will help to advance’ the Singapore Government’s Smart Urban Mobility Project, which seeks to enhance the country’s public transport systems.

“Our long-term goal is to contribute to the Singapore smart city.”

Initially, the project will deploy simulator-based software systems. After that, the plan is to move on to conducting tests on a small, unmanned vehicle, before seeking to evaluate the system on a self-driving car provided by the industry collaborators.

“Once we have such a system, we can use it to test the autonomous driving car and then report on potential issues.”

“A lot of companies are developing self-driving systems, but how can you ensure your system is robust and safe? This is our objective as we’ll be developing software to test and evaluate these systems.”

“We won’t distinguish between perception errors, planning or control errors. We just say this is a black box system and, in this project, we will open the black box.”

Back to Research@SMU May 2024 Issue