Fuzzing Autonomous Vehicles via Traffic-Rule Guided Symbolic Execution
Autonomous vehicles (AVs), often referred to as self-driving cars, have garnered significant attention and widespread adoption due to their myriad advantages. For instance, they can significantly increase fuel efficiency. Nevertheless, these AVs are not without their flaws, as evidenced by instances of unexpected braking and collisions with other vehicles or pedestrians resulting from software logic defects. Consequently, it becomes imperative to analyze AV software thorough penetration testing. However, conducting real-world tests on actual AVs poses formidable challenges, primarily due to the expensive costs involved and the paramount concern for safety. Consequently, simulation-based testing has emerged as an attractive alternative, offering a more cost-effective approach. Nonetheless, simulation-based testing is challenging in terms of generating high-quality testing scenarios. The infinite number of possible scenarios and the diverse array of real-world violations further emphasize the importance of effective scenario generation. This is where the technique of fuzzing plays a pivotal role. Previous fuzzing frameworks for AV software predominantly focused on the mutation of scenarios, with the aim of generating scenarios likely to provoke violations. For example, they changed the initial speed of the AV and the speed of obstacles. However, these frameworks often fell short in their ability to create seed scenarios—a crucial component in the mutation process. Consequently, the research team presents a systematic approach to generate seed scenarios with specific violations. The methodology begins with a meticulous examination of the driving handbook authored by the Department of Motor Vehicles (DMV). From this source, the team extracts defined violations and their corresponding driving scenarios. For instance, the vehicle cannot cross the intersection when the light is red, since it results in a red-go violation. Armed with this knowledge, the team constructs a diverse spectrum of driving sequences capable of inducing the intended violations. Subsequently, the framework transforms the generated driving graph into a seed scenario through the application of symbolic execution. Finally, the team subjects the seed scenario to fuzz testing, employing mutations across various components until the desired violation is triggered. The forthcoming evaluation will entail the application of this tool to the open-source AV software known as Apollo. Initially, the team will concentrate on scenarios occurring in specific locations or driving cases, such as intersections. It is the team's expectation that the framework will be adept at delineating multiple violations within different environmental contexts, subsequently transforming them into viable seed scenarios. Following the execution of fuzz testing, the result will comprise a comprehensive set of scenarios, each capable of instigating distinct violations.
- Record URL:
Language
- English
Project
- Status: Active
- Funding: $126079
-
Contract Numbers:
69A3552348327
-
Sponsor Organizations:
Office of the Assistant Secretary for Research and Technology
University Transportation Centers Program
Department of Transportation
Washington, DC United States 20590 -
Managing Organizations:
Center for Automated Vehicle Research with Multimodal Assured Navigation
Ohio State University
Columbus, OH United States 43210 -
Project Managers:
Kline, Robin
-
Performing Organizations:
Ohio State University Center for Automotive Research
930 Kinnear Road
Columbus, OH United States 43212 -
Principal Investigators:
Lin, Zhiqiang
- Start Date: 20231030
- Expected Completion Date: 20240830
- Actual Completion Date: 0
- USDOT Program: University Transportation Centers Program
Subject/Index Terms
- TRT Terms: Autonomous vehicles; Intersections; Performance tests; Simulation; Software; Traffic violations
- Subject Areas: Highways; Planning and Forecasting; Safety and Human Factors; Vehicles and Equipment;
Filing Info
- Accession Number: 01901380
- Record Type: Research project
- Source Agency: Center for Automated Vehicle Research with Multimodal Assured Navigation
- Contract Numbers: 69A3552348327
- Files: UTC, RIP
- Created Date: Dec 4 2023 6:46PM