Parameter-Free State Estimation Based on Kalman Filter with Attention Learning for GPS Tracking in Autonomous Driving...

Area of Science:

• Robotics and Control Systems
• Signal Processing
• Geospatial Analysis

Background:

• Maneuvering target tracking using GPS is vital for autonomous systems but challenged by complex motion and unknown sensor/noise characteristics.
• Classical Kalman filter methods struggle with parameter uncertainty and unknown color noise in GPS data, degrading performance.
• Accurate state estimation for maneuvering targets is critical for reliable navigation and autonomous vehicle operation.

Purpose of the Study:

• To develop a robust GPS-based maneuvering target localization and tracking method that overcomes limitations of classical approaches.
• To introduce a novel state estimation technique utilizing attention learning and online parameter estimation.
• To improve the accuracy and reliability of state estimation in the presence of complex dynamics and unknown GPS data characteristics.

Main Methods:

• A Kalman filter-based state estimation method incorporating attention learning via a transformer encoder and LSTM network.
• Online estimation of system model parameters using the expectation maximization (EM) algorithm, driven by attention learning outputs.
• Integration of learned system, dynamics, and measurement characteristics into the Kalman filter for state estimation.

Main Results:

• The proposed attention learning-based method demonstrated superior estimation accuracy compared to classical and pure model-free network approaches.
• Experimental validation using GPS simulation data and the Geolife Beijing vehicle GPS trajectory dataset confirmed the method's effectiveness.
• The approach successfully addressed challenges posed by complex maneuvering target motion and unknown GPS data properties.

Conclusions:

• The developed attention learning-based Kalman filter offers an effective solution for practical maneuvering target tracking applications.
• This method provides accurate state estimation without reliance on predefined system parameters, enhancing robustness.
• The findings contribute to advancing autonomous driving and navigation systems through improved GPS tracking capabilities.