Safety-Aware, Timing-Predictable and Resource-Efficient Scheduling for Autonomous Systems

Advanced driver-assistance (ADAS) and semi-autonomous systems represent the major computing demands for road vehicles, which are complex and safety-critical with strict real-time and resource constraints. With an awareness of the criticality requirements of the functions, this thesis focuses on designing shared resource scheduling methods to ensure that systems comply with criticality-related timing constraints and which improve their resilience through highly efficient resource utilisation. First, a novel graceful degradation strategy is proposed for mixed-criticality contexts based on understanding task dependency from a functional perspective to handle uncertainties (e.g., timing faults) raised by conflicts for shared resources. Compared to existing methods focusing solely on timing, the proposed strategy incorporates multiple operational modes and causality analysis-based graceful degradation. It effectively manages uncertainties and conflicts while outperforming existing methods, maximising system-wide functional Quality of Service (QoS). Second, a novel consistent mixed-criticality multi-core task static scheduling method is developed to replace the multi-system modes multi-schedules method, which can lead to unnecessary task discarding and pose challenges during system criticality mode changes. The proposed approach introduces criticality-informed temporal isolation, enables and simplifies task-level mode changes and significantly enhances the system’s resilience and the survivability of tasks, resulting in remarkable improvements in overall system performance and outperforms state-of-the-art approaches. Finally, the work is extended from the task-level to the network-level mixed-criticality data transmission. The proposed in-vehicle network scheduling method offers substantial improvements in system resilience by tolerating timing faults of safety-critical traffic, a consideration that is lacking in state-of-the-art methods. In addition, the introduced server-based method enhances bandwidth utilisation efficiency and reduces resource waste, further contributing to the overall system improvement. In summary, this thesis contributes to the resilient and efficient scheduling of shared resources in mixed-criticality systems at both the task and network levels.