Constructive Modelling of Parallelized Environmental Models for Structured Testing of Automated Driving Systems

摘要

In the automotive industry, current activities focus heavily on the development of automated driving systems (ADS). ADS process environmental data from different sensors, which are fused to generate a model of the surrounding world. Actors in the generated model are objects, which are e.g., classified as vehicles or pedestrians. The actors run in parallel, as in the real world actions from traffic participants can be taken independently and asynchronously from each other. For verification and validation of these systems a method is required, that allows for a realistic and hence parallel modeling of the system under test's environment. Additionally, the method should allow for structured testing in compliance with international norms such as the ISO 26262 and the first international standard for software testing ISO/IEC/IEEE 29119, published in 2013. In this paper we present an approach for creating environmental models for structured testing of automated driving systems with a constructive method. One step is the enumeration of all possible sequences, but we first decompose the task into manageable units by input/output dependency analysis. The expected behavior is formalized in temporal logic. In doing so, the effort for the creation of the model is feasible in industry. On the other hand, the test model guarantees the representation of all possible scenarios of use, making it a stable basis to derive significant test cases. We applied the method on an embedded system functionality in the automotive industry at AUDI. The system was architectured using the AUTOSAR 3.2 standard and implemented with Matlab Simulink. An existing, previously created test suite was available. This existing test suite served as a benchmark to assess the quality of the new test suite, derived from the environmental models. We compared the reachability of the test cases inside the implementation with code coverage measures and examined the variance of use imposed by the test suites. We present the promising results in this paper.