Enhancing GPS/IMU Localization Accuracy in Autonomous Vehicles through Deep Learning-Based Error Correction Models

Improving the precision of GPS/IMU localisation in autonomous cars is crucial for ensuring safe and efficient navigation. Several research studies have concentrated on enhancing the precision of localisation systems by employing sensor fusion and sophisticated algorithms. This study proposes a novel approach to enhance GPS/IMU localization accuracy in autonomous vehicles using deep learning-based error correction models. Leveraging Long Short-Term Memory (LSTM) networks, the method captures and corrects inherent errors in GPS/IMU data, leading to significantly improved positional accuracy. The LSTM model was meticulously designed to process sequential data, incorporating an LSTM layer with 50 units to capture temporal relationships and a Dense output layer to predict corrected longitude and latitude values. The model was trained using the Adam optimizer and mean squared error (MSE) loss function, achieving notable reductions in prediction error across 50 epochs. Comparative analyses between actual and predicted coordinates demonstrated the model's high precision. The model was further validated through deployment in a Flask application for continuous testing and a web application for real-time tracking of autonomous vehicles. Results underscore the potential of deep learning models to substantially improve localization accuracy, thereby enhancing the reliability and performance of autonomous navigation systems.