Collaborative Research: Elements: MELIOREM: An Integrated Evaluation Cyberinfrastructure towards Safe and Dependable Autonomous Driving Systems

Autonomous vehicles offer profound societal benefits, promising increased productivity and enhanced quality of life by reducing traffic congestion and improving transportation accessibility. Ensuring the safety of autonomous vehicles is paramount, given their operation on public roads and interaction with human beings. The project aims to develop MELIOREM, an automated tool designed to enhance the safety of autonomous vehicles.

By utilizing our nation's high-performance computing infrastructure, MELIOREM will conduct rigorous testing to identify and address potential safety issues before they impact public roads. This initiative ensures that autonomous vehicles are dependable and safe for all road users. Using advanced search techniques, MELIOREM will simulate various driving scenarios to assess how well these vehicles perform under different conditions, leveraging extensive computational power for complex calculations and analysis.

By bolstering the safety of self-driving technology, this project not only advances transportation safety but also provides a valuable resource to academia and industry, contributing to the broader professional community. It also creates educational opportunities by training students from diverse backgrounds in higher education.

Autonomous vehicles (AVs) promise vast societal benefits of increasing productivity and improving quality of life, from reducing traffic congestion to improving access to transportation. Ensuring AV safety is critical to success in the marketplace, and an essential aspect of AV development to ensure safety is testing. Existing techniques incorporate computerized simulation-based iterations, where the AV under evaluation is stress tested by perturbing traffic parameters and AV internal states to generate safety cases for analysis, identify AV vulnerabilities, and mitigate safety hazards.

This process largely involves using high-performance computing (HPC) infrastructure given the enormous amount of computation resources demanded by the simulations. However, current approaches often face state space explosions due to the large search spaces in both internal program executions and external environment parameters when searching for safety cases, making existing tools far from being comprehensive and efficient in HPC.

Furthermore, due to the complicated structure of AV software stack, error resilience is not yet well understood, making diagnosis and protection extremely time consuming. This project will develop an efficient and comprehensive testing infrastructure, MELIOREM, for characterizing, assessing, and identifying vulnerabilities in AV software systems in evolving traffic situations.

The core purpose of this work is practicality, enabling domain scientists to generate safety cases for characterizing and understanding AV safety, and AV developers to identify AV safety vulnerabilities using existing HPC infrastructure. This project will develop a series of algorithms to optimize test coverage, emulation efficiencies, and identify representative safety cases for an AV under test.

This work will resolve these AV development issues with respect to their practical analysis by applying MELIOREM in intelligent cyber-systems in transportation and crash analysis research domains.

This project is jointly funded by the OAC Cyberinfrastructure for Sustained Scientific Innovation (CSSI) program and the Division of Information and Intelligent Systems (IIS).

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.