A Minimum Frequency Scheduler for Autonomous Driving Systems

摘要

To cope with the complexity of autonomous driving systems, multiple modules need to be run in the same time. Nowadays the modules are run in parallel using GPU based hardware, but still there are times when only sequential access to parts of the hardware is allowed. This results in the need of developing scheduler mechanisms on top of the concurrent paradigm. The complexity of an autonomous driving system rises from the interconnection of various modules with different functionalities: ones are safety relevant, others are used for improving travel experience, while others are used for debug purposes. The safety relevant ones should be able to process all the inputs coming from the sensors, while the non-safety modules can be run less often. This paper proposes a scheduler mechanism that assures for each module a desired minimum run frequency.