A Hybrid Approach for Fault Detection

13th International Conference on Autonomous Agents and Multiagent Systems

in Autonomous Physical Agents

PID  :  119597

来源: CEUR

【 摘 要 】

One of the challenges of fault detection in the domain ofautonomous physical agents (or Robots) is the handling ofunclassified data, meaning, most data sets are not recognized asnormal or faulty. This fact makes it very challenging to usecollected data as a training set such that learning algorithmswould produce a successful fault detection model. Traditionallyunsupervised algorithms try to address this challenge. In thispaper we present a hybrid approach that combines unsupervisedand supervised methods. An unsupervised approach is utilized forclassifying a training set, and then by a standard supervisedalgorithm we build a fault detection model that is much moreaccurate than the original unsupervised approach. We showpromising results on simulated and real world domains.Categories and Subject DescriptorsI.2.9 [Artificial Intelligence]: RoboticsAutonomous vehicles,Sensors. General TermsReliability, Experimentation