With the growth of self-driving technologies, the reliability analysis of automated driving systems has received considerable attention from both academia and industry. Safety of the intended functionality (SOTIF) serves as one of the primary standards to assure the reliability and safety of the automated driving system. One of its key issues is the performance limitations of perception sensor systems. Generally, the reliability of the perception sensor system depends on the different scenarios of the driving environment. In this work, we investigate the sensor features and dependencies of the front camera and the top LiDAR of the nuTonomy scenes (nuScenes) dataset with respect to scenarios (e.g., rain and night) and distance information (e.g., two distance-based regions of interest). In addition, we apply the obtained parameters to a proven analytical reliability model to examine the impact of scenario and distance information on the reliability assessment.

Multi-sensor systems are the key components of automated driving functions. They enhance the quality of the driving experience and assisting in preventing traffic accidents. Due to the rapid evolution of sensor technologies, sensor data collection errors occur rarely. Nonetheless, according to Safety Of The Intended Functionality (SOTIF), an erroneous interpretation of the sensor data can also cause safety hazards. For example the front-camera may not understand the meaning of a traffic sign. Due to safety concerns it is essential to analyze the system reliability throughout the whole development process. In this work, we present an approach to explore the sensor’s features, such as the dependencies between successive sensor detection errors and the correlation between different sensors on the KITTI dataset quantitatively. Besides, we apply the learned parameters to a proven multi-sensor system model, which is based on Discrete-time Markov chains, to estimate the reliability of a hypothetical Stereo camera-LiDAR based sensor system.

Precise environment perception, which consists of multi-sensor systems, ensures the safety of the automated driving functions (ADFs). With the rapid evolution of sensor technologies, sensor data collection errors occur rarely. Nevertheless, accurate interpretation of the sensor data context, such as 3D multi-object tracking, is still full of challenges. Safety of the Intended Functionality (SOTIF) takes concern of the vulnerability of perception systems. The research of quantitative SOTIF analysis is still ongoing. In this paper, we propose a multi-sensor system model to observe both false negative and false positive errors in different field-of-views. Besides, we also extend a proven Markov-based approach, which takes dependencies between successive sensor errors and correlation between two dependent sensors into account, to model three correlated sensors sharing the same region of interest. In the end, we present a numerical example to illustrate the quantitative reliability analysis of a multi-sensor perception system according to SOTIF.

Modern driver assistance systems for self-driving cars often rely on data collected by different sensors to determine the necessary system decisions. To prevent system failures, different validation techniques are used. The development is often split between car manufacturers and suppliers, whereby the requested test effort is one main project acceptance criterion. Already available effort estimation methods are not applicable, because they rely on implementation details that do not exist at early phases or on project experiences or individual expert expectations, which are not reliable enough to be employed as trustworthy source. Therefore, we provide in this paper an analytic approach for the computation of the error probability of a multi-sensor system. Based on this, we can give estimations for the test effort such that with statistical confidence no errors of the sensor system can be expected during the tests. The approach is able to take both the dependencies between successive sensor errors and the correlation between different sensors into account, mainly by using discrete time Markov chains. The provided approach therefore allows to design multi-sensor systems such that a specified overall error probability can be met and to give an estimation for the upper bound of the test effort.

Modern driver assistance systems are often using a wide range of equipped sensors as primary data source. The reliability of each sensor is specified by the manufacturer and influences the system reliability distinctly. To prevent potential fatal system failures, diverse failure prevention mechanisms are included. Nevertheless, a certain level of reliability of the system has to be guaranteed to meet legal regulations. For this, extensive testing is required, which is costly due to the involved resources. To estimate and simulate the required test effort, an exact analytical method based on Markov Chains and an implementation realized in a common simulation framework is presented in this paper. It enables real automotive projects to estimate the test costs and simulate changes for various sensor setups.

The development of modern driver assistance systems in the automotive domain requires extensive testing, for safety as well as for legal reasons. This is especially the case for sensor-based systems that provide support for autonomous driving: if they fail, an accident may occur with fatal consequences. In the majority of cases, the required test effort is roughly estimated by means of previous project data, expert knowledge or based on economical factors. As an alternative, a general mathematical approach is presented in this paper, which is focused on black box systems with sensor data fusion. It enables new projects related with sensor data fusion to estimate the required test effort for a given scenario.