Quantized observer-based controller for scalable fog computing in autonomous vehicle applications

• This paper proposes a control-theory-based Scalable Fog Computing Network (SFCN) for near real-time vision data processing in autonomous vehicles using auto-scaling fog nodes. • A linear parameter-variation (LPV) model is identified with vision data as disturbance, fog nodes as control input and service-time as output. • A quantized control framework manages bandwidth limits and ensures stable data transmission. • An observer-based LPV state-feedback controller ensures ISS and robust system performance. • Simulations and real-time experiments verify the proposed control framework’s effectiveness. This paper presents the development of a control-theory-based scalable fog computing network (SFCN) designed to seamlessly process time-varying vision data offloaded by an autonomous vehicle’s (AV) onboard vision system. The AV control system requires near real-time vision data processing for navigation and application-specific tasks. The SFCN consists of a network of scalable fog nodes deployed to infer time-varying vision data within a specified service time. A quantized control framework is proposed that provides significant benefits for managing a resource-constrained SFCN while improving its vision data processing capability through auto-scaling of fog nodes. A Linear Parameter-Varying (LPV) model is developed using system identification techniques to capture the behavior of the SFCN for the specified application. Sufficient conditions for designing an auto-scaling controller are then established using input-to-state stability (ISS) theory to ensure system stability and achieve a desirable transient response while accounting for input and output quantization effects. The effectiveness of the proposed design is validated through both simulation and real-time experiments.