A reliable LSTM localization method for intelligent vehicles with GNSS outages applying chassis sensors

As the level of intelligence in vehicles continues to improve, the global navigation satellite system (GNSS) is widely used for advanced driver assistance systems and autonomous driving. It can provide a vehicle’s position in real-time, playing an important role in ensuring vehicle driving safety. However, due to obstructions such as tunnels, buildings, and trees, the GNSS signals may experience temporary outages. To accurately locate the vehicle during the GNSS outages, a reliable long short-term memory (RLSTM) localization method based on vehicle chassis sensor data is proposed. With the LSTM network, the chassis sensor data sequence is used to estimate the vehicle position. Meanwhile, a closed-loop dynamic model of the vehicle is established using the control variable sequence to predict the vehicle position. The fusion of the estimation and prediction results improves the prediction accuracy. To overcome the measurement noise of chassis sensor data, an RLSTM unit is proposed, which evaluates the reliability of chassis sensor data with control variables and only uses reliable data for vehicle localization. The proposed method is verified with actual vehicle experiments. When the GNSS outages for 3 s, the mean localization error of the proposed method decreases by 72.96% compared to the basic method. Results show that the proposed method significantly enhances vehicle localization accuracy. This method is expected to be applied to intelligent vehicles, improving driving safety.