Maneuver and Data Optimization for High Confidence Testing of Future Automotive Cyber-Physical Systems

Abstract:

 This project addresses urgent challenges in high confidence validation and verification of automotive vehicles due to on-going and anticipated introduction of advanced, connected and autonomous vehicles into mass production. Since such vehicles operate across both physical and cyber domains, faults can occur in traditional physical components, in cyber components (i.e., algorithms, processors, networks, etc.), or in both. Thus, advanced vehicles need to be tested for both physical and cyber-related fault conditions. The goal of this project is to develop theory, methods, and novel tools for generating and optimizing test trajectories and data inputs that can uncover both physical and cyber faults of future automotive vehicles. The level of vehicle reliability and safety achieved for current vehicles is remarkable considering their mass production, low cost, and wide range of operating conditions. If successful, the research advances made in this project will enable achieving similar levels of reliability and safety for future vehicles relying on advanced driver assistance technologies, connectivity and autonomy. The project will advance the field of cyber-physical systems, in general, and their lifecycle management, in particular. The validation and verification theory and methodology for cyberphysical systems will be expanded for uncovering anomalies and faults, especially using comprehensive case-based and optimization-based techniques for test scenario generation. The theoretical advances and case studies will contribute to the state-of-the-art in optimal control theory, game theory, information theory, data collection and processing, autonomous and connected vehicles, and automotive control. Sampling-based vehicle data acquisition and vehicle-aware data management strategies will be developed which can be applied more broadly, e.g., to cloud-based vehicle prognostics / conditional maintenance and mobile health-monitoring devices. The research results will be integrated into courses at the University of Michigan and a short course for industry, and they will also be disseminated through journal and conference publications. A website will be established to host down-loadable software and documentation. A variety of students at graduate and undergraduate levels, including qualified minority students, will be engaged in the project.  Several approaches will be pursued for generation and optimization of test trajectories and data inputs which maximize opportunities to discover faults. These include the development of model-free trajectory optimization techniques for actively falsifying time domain specifications that can be used directly on the experimental hardware of automotive vehicles or, alternatively, on their high-fidelity black box type models. In addition, game theory-based techniques to inform interactive in-traffic relevant trajectories for test trajectory generation will be investigated. The interactions between autonomous and human driven vehicles will be modeled and the constraints and cost (utility) functions optimized by drivers of nearby vehicles in typical traffic interaction scenarios will be characterized in order to exploit the framework of multi-move dynamic games and feedback Stackelberg equilibria to generate test trajectories representative of vehicle interactions. Furthermore, approaches to optimizing sampling-based vehicle data acquisition strategies and mining the collected trajectory data (for real world test trajectories) will be developed for uncovering faults in both existing and future vehicle fleets. Load shedding and approximate query processing techniques will be extended and adapted to a vehicular context. Approaches for efficient on-board data collection and aggregation will be implemented in a Cyber-physical system (CPS) Black Box prototype. Finally, the development of a vehicle-aware data management system (VDMS) will be pursued, leading to optimized use of data mining and compression inside the CPS Black Box to aggressively reduce the communication and computational costs. Synergistically with theoretical and methodological advances, automotive case studies will be undertaken with both realistic simulations and real experiments in collaboration with an industrial partner (AVL). These case studies will involve engine control system, a vehicle adaptive cruise control system and lateral and longitudinal maneuvers of connected autonomous vehicles with obstacle avoidance. These case studies will demonstrate and practically validate theoretical advances made.