Improved weighting in particle filters applied to precise state estimation in GNSS

Area of Science:

• Robotics and Intelligent Transportation Systems (ITS)
• Artificial Intelligence (AI) in Navigation
• Bayesian Inference and Signal Processing

Background:

• Increasing sensor fusion complexity (proprioceptive, exteroceptive, GNSS) necessitates advanced AI strategies for navigation filters.
• Bayesian inference algorithms underpin current Positioning, Navigation, and Timing (PNT) systems, but Particle Filters (PF) are computationally intensive.
• Kalman Filters (KF) have limitations with non-linear models and non-Gaussian errors, making PF desirable but often impractical due to computational cost.

Purpose of the Study:

• To present a novel methodology, Multiple Weighting (MW), that reduces the computational burden of Particle Filters (PF).
• To improve the efficiency of PF by leveraging mutual information between measurements and the unknown state.
• To assess the MW scheme's effectiveness using standalone Global Navigation Satellite System (GNSS) estimation as a baseline.

Main Methods:

• Development of the Multiple Weighting (MW) methodology to reduce computational cost in Particle Filters (PF).
• Integration of mutual information from input measurements into the PF sampling process.
• Modification of the conventional PF routine to enable more efficient posterior distribution sampling, utilizing *a-priori* state-measurement knowledge.

Main Results:

• The proposed MW strategy achieves desired accuracy levels with a significant reduction in the number of particles required.
• Standalone GNSS estimation demonstrates the method's viability as a foundation for complex multi-sensor integrated solutions.
• A considerable reduction in computational effort leads to improved state estimation accuracy, ranging from 20-40%.

Conclusions:

• The Multiple Weighting (MW) method offers a computationally efficient alternative to standard Particle Filters (PF) for navigation applications.
• MW enables enhanced accuracy in state estimation for systems like GNSS, even with limited computational resources.
• This approach provides a practical solution for achieving high-precision PNT required by ITS and robotics.