Why Track Testing Is Becoming Critical Again for Advanced ADAS and Autonomous Driving

As China accelerates the commercialization of advanced autonomous driving, limitations in existing validation frameworks—based on standardized tracks and public-road testing—are becoming increasingly apparent. These approaches are struggling to cope with the complexity of real urban environments. At the International Summit on Connected Vehicles held during Automechanika Shanghai, Professor Xiong Lu of Tongji University emphasized that track testing is once again emerging as a critical pillar in autonomous driving validation.

By Sang Min Han _ han@autoelectronics.co.kr







Displayed on the screen were numerous accident cases involving Level 2, Level 3, and Level 4 systems. Since 2023, China has been pushing forward the commercialization of high-level autonomous driving (Level 3 and above) as a national initiative. Road testing is currently underway in more than 50 cities, while pilot projects for robotaxis and unmanned vehicles are rapidly expanding.

Against this backdrop, Professor Xiong Lu pointed out that the limitations of the current multi-layered autonomous validation framework—spanning Level 2, Level 2+, Level 3, and Level 4—are becoming increasingly evident, underscoring the urgent need to enhance and rethink the role of track testing.


 

Traditionally, autonomous driving validation has followed a three-stage structure: simulation → closed-track testing → public-road testing. However, once systems reach the commercialization phase, public-road testing inevitably carries high risk and uncertainty. As a result, the industry is increasingly calling for a shift toward a two-stage framework centered on simulation and advanced track testing.

“The problem is that current tracks are far too simple to reproduce the complexity of real urban scenarios,” Professor Lu explained.
“They cannot fully replicate common but critical situations such as lane merging, side-by-side driving with heavy trucks, sudden congestion, emergency braking vehicles, or the diverse interaction patterns among road users that frequently lead to collisions and near-misses in real traffic.”


 
To address these challenges, Professor Lu’s research team has developed a cloud-based autonomous driving evaluation system. The platform integrates scenario generation, deployment, and replication technologies; digital twin–based infrastructure modeling; and large-scale participant control platforms involving dummy vehicles and pedestrians. Its goal is to quantitatively evaluate the ‘multidimensional intelligence’ of autonomous driving systems.

The system can reproduce a wide range of non-standardized test scenarios, including large dummy vehicles, moving obstacles, sudden pedestrian intrusions, and abrupt scenario changes. Crucially, it enables extreme scenario testing that would be too dangerous to conduct on public roads.

Professor Lu stressed that high intelligence alone is not sufficient for autonomous driving. In complex traffic environments, vehicles must demonstrate context-aware capabilities such as anticipation, yielding, and cooperation. Future evaluation frameworks, he argued, must incorporate these interaction-based competencies.

He added that the new evaluation system is expected to be applied not only in corporate development and testing, but also as part of future regulatory certification processes.

“Multiple government agencies and industry experts have already reviewed and validated this system, and discussions are underway to align it with international standards such as ISO,” Professor Lu noted.
“The real problems of autonomous driving emerge on the road. If that is the case, closed tracks must become intelligent enough to truly replace the road itself.”

As China continues to accelerate autonomous driving regulation and commercialization, innovation in track testing is becoming the next battleground—one that will ultimately determine the safety, credibility, and public trust in autonomous mobility across the industry.

AEM(오토모티브일렉트로닉스매거진)

---