Enhancing IMU Navigation Systems with LSTM and ZUPT: A Machine Learning Approach

This paper presents a cutting-edge approach to eliminating errors in Inertial Measurement Unit (IMU) systems and enhancing trajectory positioning using machine learning techniques. By remotely acquiring IMU data through a Raspberry Pi, the system utilizes a Long Short-Term Memory (LSTM) network to reduce errors. It then employs Zero Velocity Update (ZUPT) to estimate speed and plot trajectories. Our methodology achieves significant reductions in IMU error, with a Root Mean Square Error (RMSE) below 0.08, and maintains a positioning accuracy around 2%. The results demonstrate the efficacy of neural network learning in precise trajectory tracking and positioning, highlighting the transformative potential of machine learning in IMU-based navigation technologies.