Fault Detection Using Sensors Data Trends for Autonomous Robotic Mapping

Abstract

Map building of an environment is the foremost need for precise automatic robotic navigation. Moreover, fault detection and recovery are importantly required for every fully automatic system. The presented work describes influence of qualitative trends of the data streams coming from IMU sensor, Odometry senso, and Mapping in order to detect faulty creation of the map of an environment at runtime. It detects qualitative trends (inc, dec, con) within sensors data using a special software entity called Binary Qualitative Observer (BiQObs). BiQObs takes two data streams from robot sensors, extracts their qualitative trends, and finally compares these trends to find out whether they match or not. The work comprises of basically three main activity phases: Observe, Detect Fault and Repair Fault. First phase Observe uses BiQObs for generating and observing trends in sensors data specifically YAW measurement of the robotic pose from three different sources namely IMU, odometry, and pose from mapping. Besides BiQObs, two more observers namely Node Observer (NObs) and General observer (GObs) are also used for monitoring system nodes and their topics. Second phase Detect Fault uses Diagnosis Model Server and Diagnosis Engine to detect the root cause of the fault caused by mismatch between qualitative trends. Last phase Repair Fault uses Diagnosis Repair Engine and action servers for repairing the diagnosed fault. Finally experimental results are evaluated using the scenario of mapping of an unknown environment with Pioneer-3DX robot equipped with SICK laser and IMU sensors. Evaluation process successfully detects fault like incorrect odometry due to slippage of wheels, and brings the case in normal situation in a remarkable time of 13.54 seconds.