Identification of Hazardous Driving Scenarios Using Cross-Channel Safety Performance Indicators
conference paper
Automated Driving (AD) vehicles are slowly being deployed on public roads. These AD vehicles will encounter hazardous (dangerous) scenarios due to unforeseen edge cases at design time and changing environments on the road after deployment. To allow developers of AD systems to mitigate such unforeseen risks, the safety of AD vehicles needs to be continuously monitored after deployment. To this end, the UL4600 standard and AVSC guidelines recommend the use of safety performance indicators (SPIs) by AD vehicle developers. Our paper presents a framework that uses SPIs to identify potentially hazardous scenarios specific to the evaluated AD system, covering both AD vehicles and cloud operations. The framework uses the perception systems and motion planners of heterogeneous redundant multi-channel architectures to detect hazards invisible in single-channel-based systems, provided one of the channels observes the environment correctly. We propose three cross-channel SPIs and use them to identify hazardous scenarios in the AD vehicle and validate this approach with a proof-of-concept implementation. In a test of 6 challenging routes in the CARLA simulator, our framework automatically identifies 86% of hazardous situations. Next, it identifies contributing issues in the AD vehicle, such as missed object detections or dangerous planned trajectories. With this proof of concept, we show that this framework provides evidence on the safety of deployed systems, identifies AD vehicle functions in need of improvement and provides lessons for the development of future AD systems.
Topics
TNO Identifier
1014926
ISSN
15301591
ISBN
978-398267410-0-0
Publisher
Institute of Electrical and Electronics Engineers Inc (IEEE)
Source title
Proceedings -Design, Automation and Test in Europe, DATE
Collation
7 p.
Files
To receive the publication files, please send an e-mail request to TNO Repository.