Service Orchestration for Object Detection on Edge and Cloud in Dependable Industrial Vehicles

Abstract

Industrial applications, including autonomous systems and vehicles, rely on processing data on multiple physical devices. The composition of functionality across heterogeneous computing infrastructure is challenging, and will likely get even more challenging in the future as software in vehicles is updated to introduce new features and ensure the safety. New soft real-time use cases emerge and in such cases the model of offloading processing from a limited or malfunctioning device is a viable solution. This study examines orchestration of services across edge and cloud for an industrial vehicle application use case involving image based object detection using machine learning (ML) based models. First, service orchestration requirements are defined taking into account the dependable nature of industrial vehicle applications. Second, an implementation based on Arrowhead framework is presented and evaluated. The open Arrowhead framework offers means for dynamic service discovery, authorization and late binding of computational units. The feasibility of object detection as a service and the suitability of Arrowhead framework to support such orchestrations across edge and cloud is assessed.