IMU Data and GPS Position Information Direct Fusion Based on LSTM

Area of Science:

• Deep learning applications in navigation systems.
• Advancements in inertial navigation technology.

Background:

• Traditional inertial navigation systems (INS) suffer from cumulative errors.
• Integrating Global Positioning System (GPS) with INS can mitigate drift, but optimal fusion remains a challenge.
• Deep learning offers novel approaches to enhance navigation accuracy.

Purpose of the Study:

• To propose and evaluate a Long Short-Term Memory (LSTM) based method for position estimation.
• To fuse data from Inertial Measurement Units (IMU) and GPS using LSTM.
• To compare the performance of the LSTM method against traditional Kalman Filter (KF) based SINS/GPS systems.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) networks for data fusion.
• Employed Inertial Measurement Unit (IMU) and Global Positioning System (GPS) data.
• Conducted simulations and experiments in both static and dynamic scenarios.
• Explored hyperparameter ranges for LSTM models.
• Compared results with Strapdown Inertial Navigation System (SINS)/GPS integrated navigation systems using Kalman Filter (KF).

Main Results:

• LSTM method demonstrated reduced position error Standard Deviation (STD) compared to SINS.
• In simulations, LSTM position error STD was 52.38% of SINS.
• In experiments, LSTM position error STD was 23.08% using only SINS data.
• Maximum radial errors in simulations and experiments were 0.57 m and 1.31 m, respectively.
• LSTM fusion method showed no cumulative divergence error.

Conclusions:

• The trained LSTM model is a dependable method for fusing IMU and GPS data.
• The proposed LSTM approach enhances position estimation accuracy in inertial navigation.
• LSTM offers a viable alternative to traditional methods for integrated navigation systems.