Machine learning-based multi-sensor fusion for warehouse robot in GPS-denied environment

Mobile robots have been widely used in warehouse applications because of their ability to move and handle heavy loads. This study deals with sensor fusion of Inertial Measurement Unit (IMU) and Ultra-Wide Band (UWB) devices like Pozyx for indoor localization in a warehouse environment. UWB is a key positioning technology for the complex indoor environment and provides low-cost solutions for sensor fusion. The IMU and position data are collected for different random trajectories from Pozyx tags placed in the Turtlebot2i differential drive mobile robot. These data are used to estimate the position and orientation of the robot. Different filters like Long Short Term Memory (LSTM), Convolution Neural Network based LSTM (CNN-LSTM), Multi-Layer Perceptron (MLP), and Convolution Neural Network (CNN) filters are employed for accurate indoor positioning. Furthermore, the performance of IMU sensor fusion with IMU + UWB sensor fusion was compared based on Mean Absolute Error (MAE), Loss, and Root Mean Square Error (RMSE) for different trajectories. Simulations and experimental tests were carried out for different trajectories and the test results show that fused data using CNN-LSTM had less positioning error compared to other filters.