This volume of the OSADL Academic Works series presents a thesis that examines the viability of statistical methods to cope with the safety requirements of next-generation complex autonomous systems.

To meet the requirements of complex autonomous systems it may be advantageous to deploy a fully grown operating system such as Linux that can provide reliability, performance, security and updates. However, due to its resource-sharing architecture, traditional safety-related verification processes may no longer be feasible or, if so, not solely sufficient. The work presented herein proposes to complement the traditional approach with statistical analysis to pave the way towards the certification of safety-related complex applications. Specifically, it contributes a novel statistical analysis technique to quantify the execution path coverage of the Linux kernel and to estimate the residual risk resulting from untested execution paths.

Starting by examining the main gaps in the field of test coverage with respect to the Linux kernel, the work goes on to statistically estimate the current test coverage by analyzing the execution paths traversed during a testing campaign. The proposed methods are demonstrated on the example of an autonomous emergency braking system that is based on a Linux kernel in combination with machine learning and is, as such, representative of next-generation safety-related systems. On the basis of this case study the inherent non-determinism of the Linux kernel is revealed, and it is shown that estimating test coverage with the proposed statistical methods is viable. Finally, a technique to quantify the testing process and the risk associated with uncovered paths is presented.