CRII: CNS: Bringing Predictable Real-time Computing to Connected Autonomous Driving Systems

点击翻译按钮获取中文摘要

Connected vehicle technology is a promising solution to provide reliable autonomous driving that will change the traditional transportation system by building stable, interactive wireless communications between vehicles, the smart infrastructures (e.g., the roadside unit), and personal communications devices. However, achieving reliable and safe connected autonomous driving (CAD) is still very challenging. On one hand, the safety of the CAD system hinges critically on its timing correctness, as crucial driving decisions fully depend on the output of the real-time perception system. On the other hand, requesting information from other devices mayl create additional delays for the on-vehicle real-time perception tasks, and thus the timing correctness of the CAD system can be easily violated by unpredictable communications. This project seeks to bring predictable real-time computing to CAD systems, and the goal of the proposed research is to enable the connected autonomous vehicle and exterior devices to perform real-time perception tasks as a whole by (i) establishing a practical real-time task model to integrate exterior devices into the on-vehicle perception system, which can be implemented on the GPU-enabled computing platforms; (ii) proposing real-time task scheduling algorithms and associated timing validation analysis to guarantee that all the real-time perception tasks can complete at the right time; (iii) developing a prototype CAD system on the autonomous vehicle testbed, HydraOne, and the roadside unit, Equinox, to evaluate the real-time performance of the proposed solutions.Building a CAD system will constitute a major technological breakthrough towards realizing fully autonomous vehicles. In particular, this project emphasizes both scheduling algorithm design and system implementation. The establishment of a real-time suspending-gang task model will enable the first-of-its-kind formalization for depicting the executing flow of real-time workloads executed between the autonomous vehicle and the exterior devices. The real-time task scheduler oversees the entire system and ensures its timing correctness. The creation of new real-time resource allocation methods together with the associated analysis for validating timing constraints will drive the scheduling theory towards real applications in future cyber-physical systems. The proposed research aims to realize the CAD system on the physical platforms (HydraOne/Equinox), with indoor and outdoor studies beyond simulation. Especially, HydraOne/Equinox are ready-to-use platforms that will allow experts/researchers to easily examine their research designs regarding autonomous driving. Educational efforts will be devoted to (i) develop the HydraOne Educational Toolkit for undergraduate education and research, (ii) curriculum design for hands-on learning in the BS/MS program, (iii) summer camp development for K-12 students and teachers, (iv) broadening participation in computing and engineering to enhance diversity.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.