Ground Classification for Robots Navigation Using Time Series Dataset with LSTM and CNN

This paper focuses on enhancing the navigational capabilities of autonomous robots through the prediction of surface types using time series data from Inertial Measurement Units (IMU) sensors. This development is essential for advancing fully autonomous robots that can efficiently adapt to different surface conditions. The dataset was collected from a mobile robot while traversing various surfaces such as concrete, carpet, tiles … etc. Long Short-Term Memory and Convolutional Neural Network models were employed to classify surface types based on features like acceleration and velocity. Results indicated that the CNN-BiLSTM model significantly improved prediction accuracy, achieving an accuracy of 80.05%.