The introduction of fully autonomous vehicles posesa number of concerns regarding the safety and dependability ofvehicle operation. Best practice standards within the automotiveindustry rely on the driver operating the vehicle. With thetransition away from manual control, an increased emphasishas to be placed on verification during the vehicle developmentstages. The work presented within this paper aims to establisha framework for the various verification activities performedduring development, and their impact on the safety of the vehicle, as well as a set of guidelines for verification of the decision makingprocess of autonomous vehicles.
Safer autonomous navigation might be challenging if there is a failure in sensing system. Robust classifier algorithm irrespective of camera position, view angles, and environmental condition of an autonomous vehicle including different size & type (Car, Bus, Truck, etc.) can safely regulate the vehicle control. As training data play a crucial role in robust classification of traffic signs, an effective augmentation technique enriching the model capacity to withstand variations in urban environment is required. In this paper, a framework to identify model weakness and targeted augmentation methodology is presented. Based on off-line behavior identification, exact limitation of a Convolutional Neural Network (CNN) model is estimated to augment only those challenge levels necessary for improved classifier robustness. Predictive Augmentation (PA) and Predictive Multiple Augmentation (PMA) methods are proposed to adapt the model based on acquired challenges with a high numerical value of confidence. We validated our framework on two different training datasets and with 5 generated test groups containing varying levels of challenge (simple to extreme). The results show impressive improvement by$$\approx $$ 5–20% in overall classification accuracy thereby keeping their high confidence.
Fully automated vehicles will require new functionalities for perception, navigation and decision making - an Autonomous Driving Intelligence (ADI). We consider architectural cases for such functionalities and investigate how they integrate with legacy platforms. The cases range from a robot replacing the driver - with entire reuse of existing vehicle platforms, to a clean-slate design. Focusing on Heavy Commercial Vehicles (HCVs), we assess these cases from the perspectives of business, safety, dependability, verification, and realization. The original contributions of this paper are the classification of the architectural cases themselves and the analysis that follows. The analysis reveals that although full reuse of vehicle platforms is appealing, it will require explicitly dealing with the accidental complexity of the legacy platforms, including adding corresponding diagnostics and error handling to the ADI. The current fail-safe design of the platform will also tend to limit availability. Allowing changes to the platforms, will enable more optimized designs and fault-operational behaviour, but will require initial higher development cost and specific emphasis on partitioning and control to limit the influences of safety requirements. For all cases, the design and verification of the ADI will pose a grand challenge and relate to the evolution of the regulatory framework including safety standards.
To manage the complexity of C programs, architecture models are used as high-level descriptions, allowing developers to understand, assess, and manage the C programs without having to understand the intricate complexity of the code implementations. However, for the architecture models to serve their purpose, they must be accurate representations of the C programs. In order to support creating accurate architecture models, the present paper presents a mapping from the domain of sequential non-recursive C programs to a domain of formal architecture models, each being a hierarchy of components with well-defined interfaces. The hierarchically organized components and their interfaces, which capture both data and function call dependencies, are shown to both enable high-level assessment and analysis of the C program and provide a foundation for organizing and expressing specifications for compositional verification.