Online detection of calibration errors in humanoid robots

One of the reasons why humanoid robots fall is the presence of calibration errors at their joints. In this paper, we propose a method for detecting these calibration errors during online operation. We use a signal classification technique to detect differences in the signal patterns of sensors mounted on humanoids caused by calibration errors. Moreover, we present a method using ℓ₁- and ℓ₂-SVM for selecting efficient features for pattern classification. Our analysis results show that our approach correctly classified 84% of the data into normal and abnormal classes.