Machine learning based classification of slip-stick phenomena in mobile robot using IMU measurements: a comparative study of SVM, Random Forest and KNN classifiers

Abstract

Four-wheel steering mobile robots exhibit complex non-linear dynamics due to the following reasons such as steering front and rear wheel simultaneously, load transfer during turning and tyre-road friction. Traditional stability indicators mostly depend on low-frequency vehicle states or quantitative measures, which often fail to detect early signs of instability that occur due to slip. Detecting stick and slip in four-wheel steering mobile robots will support accurate positioning and precise control of mobile robots in the desired manner. This study focuses on slip-stick classification in four-wheel steering mobile robots. The study utilizes linear acceleration and angular velocity data from MPU6050 sensor. The data collected are labeled, windowed and filtered using Kalman filter to remove noise. Subsequently applying Discrete wavelet transform (DWT) on the signal to obtain time-frequency information. Energy and Entropy features are extracted from the time – frequency information followed by applying principal component analysis for data dimension reduction. The artificial intelligence (AI) algorithms applied in this study are K- Nearest Neighbor (KNN), Support Vector Machine (SVM), and Random Forest (RF) to classify slip-stick phenomena. The results indicate that the classification is better while using support vector machines classifier compared to the other learning algorithms with a mean classification accuracy of 96.59 % obtained for SVM classifier. In comparison the SVM classifier performed well in classifying the slip-stick phenomena more accurately. In future more sophisticated AI algorithms can be utilized to enhance classification followed by implementation on a more sophisticated mobile robot platform and integrating other sensors would be more effective and beneficial.