I. Introduction
Automotive systems usually have deep processing pipelines that span over multiple stages processed by different soft-ware/hardware components with data dependencies. For ex-ample, in autonomous vehicles, obstacle avoidance is real-ized by a chain of computational tasks including sensing, perception, planning and control, as shown in Fig. 1.
The example is quoted from RTSS 2021Industry Challenge, which is implemented by the company PerceptIn in autonomous driving systems.
The system must comply with timing constraints in several aspects to guarantee that the final control command outputs can be executed correctly and timely. Satisfying the end-to-end timing constraints of control paths is a prerequisite for correct and safe systems, e.g., the success of obstacle avoidance requires that the task chain finishes before predefined deadline. Violating timing constraints may lead to catastrophic consequences such as loss of human life. As a consequence, formal modeling and analysis must be performed to guarantee that the timing constraints are always honored at run-time. For this purpose, the end-to-end timing constraints have been extensively studied in the context of automotive systems [1]–[5], known as the so-called maximum reaction time, presenting the length of time interval from a stimulus to its corresponding response, and the maximum data age, describing the freshness of the data.