Safety Testing of Automated Driving Systems: A Literature Review

Abstract

The advancement of automation in safety-critical systems has opened the door to newer opportunities in several fields. However, the increasing complexity has led to more risks and the need for better safety assessment. Safety testing of Automated Driving Systems (ADS) became an important research topic due to the potential benefits of reducing traffic congestion, time, and even accidents; however, safety remains an ongoing issue. We analyzed the existing literature from the past twenty-five years by performing backward snowballing to identify the proposed safety testing approaches for ADS. From 44 selected publications, we identified safety features, testing methods, techniques, tools, and datasets used for testing an ADS, considering both multi-module and end-to-end driving systems. The community is mostly focused on multi-module driving systems, working on units of functionality related to collision avoidance, lane keeping, and warnings, testing the vehicles on-road and in simulated environments with different driving scenarios, combining different methods and techniques, and proposing custom solutions. A major research gap seems to be the lack of safety testing approaches that integrate the advantages of on-road testing with simulation testing. For future research, this study provides an introductory overview of the field for researchers and practitioners interested in ADS safety testing.

The Effects of Speed and Delays on Test-Time Performance of End-to-End Self-Driving

“…Practically, the results demonstrated that models trained at specific speeds exhibit decreased performance when tested at novel speeds. This poses a challenge for the real-life testing of end-to-end ADS because deploying models at human-like speeds increases risks [ 36 ].…”