Dynamic and execution views to improve validation, testing, and optimization of autonomous driving software

Miguel Alcon, Hamid Tabani, Jaume Abella, Francisco J. Cazorla

Abstract: The adoption of autonomous driving (AD) software executed on high-performance multi-processor systems on chip (MPSoCs) contributes to increasing the overall system’s safety and efficiency. However, existing AD software frameworks are provided as complete implementations that do not follow any domain-specific safety-requirement centric development process. In this paper, we develop, for the first time, ISO 26262 dynamic views of a representative AD framework, Apollo. Dynamic views are a key element of software architectural design that links safety software requirements with their implementation, and are the basis to verify that all casuistics are properly considered in the design and tested in the validation tests. We also show that dynamic views miss key information of the execution parallelism of Apollo, needed to assess and improve execution efficiency to meet performance-related safety requirements and reduce resource utilization. We cover this gap by proposing execution views that capture the parallelism exploited by the analyzed application on the target MPSoC. Execution views improve greatly resource usage testing, which is required by ISO 26262, and allow better resource utilization contributing to the stringent cost-reduction requirements in automotive domains.