In fact, techniques explored at our research group have proven this approach to be promising, and other work has shown that machine learning can help speed up the schedulability analysis process in certain contexts. We envision the usage of machine learning techniques to have the potential to provide fast and accurate predictions on the schedulability of task sets, even if at the expense of false positives. Using such methods can become complex and computation heavy for larger dynamic task sets, especially if additional constraints are taken into account, such as the ones introduced by task dependencies and shared resources. ![]() ![]() This idea is motivated by the disadvantages of traditional methods, which can usually provide a fail-safe analysis of the feasibility of a schedule, but are often pessimistic and can lead to recurrent calculations. We consider that task migration planning can be enabled by using machine learning techniques for supporting the prediction on whether task sets distributed to each ECU are schedulable. Some of these challenges are maximizing the efficiency of the system, ensuring all tasks get to execute, mapping the tasks to the devices, and ensuring that all tasks meet their deadlines, especially ones that are highly critical. ![]() However, the global implementation of this strategy for all computational tasks executed in a vehicle faces other challenges, especially when considering the diversity of the tasks, ranging from safety-critical ones to entertainment and comfort functions. This solution might seem obvious, and it is actually starting to be implemented in industry for some vehicle functionalities. To counter this, ECU consolidation is pursued in the industry, with the aim of moving the execution of tasks to a few powerful, multiple-purpose devices. Nonetheless, if this number were to keep growing, this would soon turn into an unsustainable development, as it presents efficiency problems when considering the power consumption, cost, and weight of the car. As a result, dozens of electronic control units (ECUs) can be found in modern cars. Until recently, the trend was to add many single-purpose devices for new tasks in the vehicle. Our approach shows a promising potential for machine-learning-based schedulability analysis and enables a comparison between different ML models. To demonstrate the capabilities of the setup, we show its integration with FreeRTOS-based ECUs and two ML models-a long short-term memory (LSTM) network and a spiking neural network-along with a collection of tasks to distribute among the ECUs. The migration is aided by the machine learning predictions on the schedulability analysis of possible future task distributions. ![]() As part of a holistic testbed, we introduce a collection of reproducible tasks, as well as a toolchain that controls the dynamic migration of tasks depending on ECU status and load. In this paper, we present a setup with a generalistic and modular architecture that allows for integrating and testing different ECU architectures and machine learning (ML) models. In our research, we aim to enable ECU consolidation by migrating tasks at runtime between different ECUs, which adds redundancy and fail-safety capabilities to the system. Consolidating tasks to a smaller number of electronic control units (ECUs) is an important strategy for optimizing costs and resources in the automotive industry.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |