Slip Ratio Prediction in Autonomous Wheeled Robot using ROS-Physics Engine based Hybrid Classification Approaches

This study presents an approach for predicting the slip ratio, a significant parameter in improving the autonomous navigation of robots. The novelty of the work is presented in two parts; first is the design of a Robot Operating System (ROS)-Physics Engine-based hybrid machine learning model with separate terrain classification and slip ratio estimation algorithms. The motivation for developing a hybrid model is presented by analyzing slip ratio-longitudinal friction coefficient characteristics in three different terrains: sand, grass, and asphalt. The longitudinal friction coefficient ( μ x ) values are distinct for the same values of slip ratio in different terrains, potentially influencing the model’s feature extraction. The features for the slip estimation model are collected from torque sensors and IMU, which measure the external forces acting on robot wheels. The slip estimation is achieved using ensembles of weak classifiers to reduce the computational load of onboard systems. An optimized model with slip ratio classes as output has been evaluated and presented for three different terrains. The model provides an accuracy of 78.1% in the sand, 77.8% in the grass, and 73.5% in asphalt in simulations. The model is compared with a strong classifier used in previous works regarding the accuracy, training time, and computational time. Second, the virtually- trained learning algorithm is tested and validated using the physical robot whose virtual model is implemented in simulations with the actual slip ratio estimated using an experimental framework.