Abstract— This paper presents how execution monitoring can be applied to detect sensor measurements faults and shows how to use this information to improve data estimation processes. Experimental results based on ultrasonic sensors measurement data are presented and the estimation process improvement could be verified. I. INTRODUCTION n order to attain a good performance, a control system must be based on a suitable model of the process to be controlled, and need to be fed with confident measurements of the state variables of interest. If the sensory data, used to calculate the system’s state variables, are both noisy and corrupted, even a stable and well designed controller may present a deteriorated performance. Depending on its application, such performance deterioration may cause a little damage or even a big disaster. Data estimation or data fusion processes are commonly used to improve sensory data confidence. Among the several estimation and data fusion techniques available, those based on the distinct forms of the Kalman Filter seems to be preferred. According to [8], such preference may be explained by the fact that the Kalman Filter is an optimal estimator, in the sense that it minimizes the mean squared error, when it is possible to assume that the system model noise and the measurement or observation error are Gaussian, zero-mean sequences [2], [11]. However, such assumption is no longer supported if a sensor fails or saturates. This may be the case when ultrasonic sensors measurements are the observations of an Extended Information Filter, used to estimate the robot pose with respect to a corridor centerline or to a reference wall [10], [12]. In such application the ultrasonic sensors used failed or Manuscript received June 8, 2008. This work was supported in part by CENPES/PETROBRAS and CNPq. Eduardo O. Freire is with the Electrical Engineering Nucleus of Federal University of Sergipe – NEL/UFS, Av. Marechal Rondon, S/N, São Cristóvão-SE, Brazil, 49100-000 (phone: +55-79-2105-6834; fax: +55-79- 2105-6684; e-mail: efreire@ufs.br). Elyson A. N. Carvalho is with the Electrical Engineering Nucleus of Federal University of Sergipe – NEL/UFS, Av. Marechal Rondon, S/N, São Cristóvão-SE, Brazil, 49100-000. He is also a PhD. student at the Electrical Engineering Departament of the Federal University of Campina Grande – DEE/UFCG, Av. Aprígio Veloso, 882, Bodocongó, Campina Grande-PB, Brazil, 58109-900 (e-mail: ecarvalho@ufs.br). Carlos A. V. Cardoso is with the Electrical Engineering Nucleus of Federal University of Sergipe – NEL/UFS, Av. Marechal Rondon, S/N, São Cristóvão-SE, Brazil, 49100-000 (e-mail: cvcardoso@ufs.br). Benedito A. Luciano is with the Electrical Engineering Departament of the Federal University of Campina Grande – DEE/UFCG, Av. Aprígio Veloso, 882, Bodocongó, Campina Grande-PB, Brazil, 58109-900 (e-mail: benedito@dee.ufcg.edu.br). reached its saturation level due to the characteristics of the operation environment, like opened doors along the corridor, or the bad orientation of the robot with respect to the side walls, and the use of additional sensors or measurements would not be helpful to improve the confidence of the robot’s pose estimated by the Extended Information Filter. This work formalizes the use of execution monitoring to detect sensor measurement faults in order to improve data estimation processes. As a case of study, the proposed method is applied to detect ultrasonic sensor measurement faults, like biased measurement errors and saturated measurements, allowing the use of an Extended Information Filter (EIF) to estimate the distance between the ultrasonic sensor and the detected obstacles. According to [8], “execution monitoring is a continuous real-time task of determining the conditions of a physical system, by recording information, recognizing and indicating anomalies in the behavior.” A monitoring system should be able to detect, isolate and identify faults. Fault isolation means the ability to diagnose which part of the system failed, whereas fault identification consists in determine the magnitude of the fault [8]. Execution monitoring is widely applied in industrial control and automation, most known in this research area as FDI – Fault Detection and Isolation – but it is not very used in robotics [8]. Despite of this, [8] presents a wide survey about the application of execution monitoring in robotics, particularly focused on autonomous mobile robots. Execution monitoring systems may be classified to one or more of the following three approaches: analytical, data- driven, and knowledge-based [7], as shown in Fig. 1. The execution monitoring system proposed in this paper to perform the detection of faults in ultrasonic sensors measurements can be classified as analytical, more precisely, as an Observer. This paper is organized as follows. In Section 2 the problem definition is presented. Section 3 is about the EIF. The proposed observer as an execution monitoring system to perform sensory fault detection is presented in Section 4. The obtained results are shown in Section 5. Conclusions and propositions for future are presented in Section 6. II. PROBLEM DEFINITION In [12] a corridor navigation and wall-following stable controller for sonar-based mobile robots was presented. Sonar measurements may deteriorate or be impossible to obtain under certain circumstances as, for example, when the robot is travelling close to an open door in the corridor, or when the robot has a significant angle of deviation from the Execution Monitoring Applied to Data Estimation Processes Eduardo O. Freire, Elyson A. N. Carvalho, Carlos A. V. Cardoso, and Benedito A. Luciano I