Distributed to embedded Bayesian Network for diagnosis of a networked robot

Abstract

Mobile networked robots are distributed systems controlled by a distant station i.e, the controller is implemented on a control station. The main mission of the mobile robot is reaching a target position starting from an initial one while receiving the instructions from the control station via a wireless network. The wireless network is characterized by a stochastic behavior and is sensitive to perturbations. The unreliability of wireless networks does not guarantee data transmission between system components (the robot and the station) which can cause system performance degradation. A distributed Bayesian Network (BN) was proposed in previous work to monitor and diagnose the system performance and to model causal uncertainties between failures. The developed BN is a modular Bayesian Network (MBN) which is composed of three Bayesian modules shared between the robot and the control station. In the case of system performance degradation because of a bad network state, the robot switches to embedded controller implemented on-board (autonomous behavior). The distributed diagnosis architecture must be updated and the diagnosis tool becomes embedded on the robot. A procedure of BNs assembly is described in this work in order to implement one monolithic BN on-board. The obtained monolithic BN is the result of combining two Bayesian modules from the modular Bayesian Network: the control and the operative Bayesian modules.