Adaptive predictive control for pipelined multiprocessor image-based control systems considering workload variations

Image-based control (IBC) systems have a long sensing delay. The advent of multiprocessor platforms helps to cope with this delay by pipelining of the sensing task. However, existing pipelined IBC system designs are based on linear timeinvariant models and do not consider constraint satisfaction, system nonlinearities, workload variations and/or given interframe dependencies which are crucial for practical implementation. A pipelined IBC system implementation using a model predictive control (MPC) approach that can address these limitations making a step forward towards real-life adaptation is thus promising. We present an adaptive MPC formulation based on linear parameter-varying input/output models for a pipelined implementation of IBC systems. The proposed method maximizes quality-of-control by taking into account workload variations in the image processing for individual pipes in the sensing pipeline in order to exploit the latest measurements, besides explicitly considering given inter-frame dependencies, system nonlinearities and constraints on system variables. The practical benefits are highlighted through simulations using vision-based vehicle lateral control as a case study.