Graceful Degradation with Condition- and Inference-Awareness for Mixed-Criticality Scheduling in Autonomous Systems

Abstract

In an autonomous system, understanding functional causality is crucial when designing a shared resource scheduling strategy. Without this, the system would be unlikely to achieve a holistic functional quality of service (QoS) and may not meet safety goals. This work proposes a novel graceful degradation strategy in a mixed-criticality context. Instead of discarding computational load at the application level, a qualitative and quantitative definition of importance order is used to assist in realizing finer-grained task-level degradation and preserve more LO-criticality tasks. The causality analysis-based degradation bridges the gap where functional dependencies are not considered in the scheduling design and, thus, leads to the system being able to continue to run with relatively high QoS during the degradation process.