Early Terrain Identification for Mobile Robots Using Inertial Measurement Sensors and Machine Learning Techniques

Abstract

Due to rapid advancements in robotics technology, mobile robots are now utilized across various industries and applications. Understanding the terrain on which a robot operates can greatly aid its navigation and movement adjustments, ultimately minimizing potential hazards and ensuring seamless operation. This study aims to identify the specific terrain on which a mobile robot travel. Data was gathered using an inertial measurement unit (IMU) installed on the robot for experimental testing. The key contributions of this research are twofold: firstly, the implementation and evaluation of various machine learning techniques using the IMU sensor dataset, comparing their performance using metrics like accuracy, precision, recall, and F1-score. Secondly, after assessing the different techniques, the most effective one is chosen for the final system implementation. Following the experimental evaluation of machine learning techniques, it was determined that the light gradient boosting machine (LGBM) classifier outperformed the others. Consequently, LGBM was utilized for the proposed system's implementation, achieving a 91% accuracy in surface classification. The experimental results highlight the efficiency and viability of the proposed system.