Outlier Rejection for Networked Control Systems based on Middleware Conference Paper uri icon

abstract

  • 2017 IEEE. In cyber-physical systems, state estimation errors can be caused not just by process noise or measurement noise in sensors, but also by errors in the communication network. Jitter in packet delivery over a communication network carrying sensor measurements can result in timing errors which result in large outlier type state estimation errors, such as, for example, velocity estimation errors in vehicular control networks. These large outlier types of errors can seriously affect performance and robustness of the overall system. For reliability of cyber-physical systems, it is important to design them to be robust to such types of outliers. We propose a two-pronged approach involving a flexible adaptive model identification algorithm with outlier rejection, which in turn uses an adaptive system model to detect and reject outliers, thus shielding the estimation algorithm and thereby improving reliability. We show that the outlier rejection approach which intercepts and filters the data, combined with simultaneous model adaptation, can result in significantly improved performance of Model Predictive Control (MPC). We demonstrate this by reducing trajectory deviation errors in a vehicular testbed. We implement the overall system over Etherware, a middleware supporting a flexible mechanism that allows the suggested algorithm not only to be remotely incorporated without additional computational burden to the downstream controller, but also to be attached or removed at runtime, without reconfiguration, to adapt to any change of the dynamic plant.

name of conference

  • 2017 14th IEEE Annual Consumer Communications & Networking Conference (CCNC)

published proceedings

  • 2017 14TH IEEE ANNUAL CONSUMER COMMUNICATIONS & NETWORKING CONFERENCE (CCNC)

author list (cited authors)

  • Ko, W., & Kumar, P. R.

citation count

  • 0

complete list of authors

  • Ko, Woo-Hyun||Kumar, PR

publication date

  • January 2017