Fault diagnosis of mobile robot based on short-term prediction using neural network

This paper discusses fault diagnosis that is estimated abnormal condition before breakdown of a mobile robot. We apply the module structured neural network to estimate abnormal conditions. Each neural network learns the dynamics model of a mobile robot in specific situation. A NN predicts the short term perceptual information, the difference with real information indicates a possibility of abnormal condition. Through experiment which compared normal condition with abnormal condition, we show that the transition of NN modules are indicated the abnormal situation.