A Nonlinear Framework of Delayed Particle Smoothing Method for Vehicle Localization under Non-Gaussian Environment

Area of Science:

• Robotics and Control Systems
• Signal Processing
• Statistical Inference

Background:

• Accurate vehicle state estimation (VSE) is crucial for autonomous systems.
• Traditional methods struggle with non-Gaussian measurement and process noises.
• Existing filtering and smoothing techniques often fall short in complex noise environments.

Purpose of the Study:

• To introduce a novel nonlinear smoothing framework, the non-Gaussian delayed particle smoother (nGDPS).
• To enhance VSE accuracy by explicitly addressing non-Gaussianity in noise models.
• To provide a robust solution for real-world VSE challenges using low-cost sensors.

Main Methods:

• Adoption of the multivariate Student's t-distribution for non-Gaussian noise PDF computation.
• Development of an Ensemble Kalman Filter (EnKF)-based approach for mean and covariance estimation.
• Implementation of a delayed Gibbs sampling algorithm to mitigate particle degeneracy and incorporate smoothing.

Main Results:

• The nGDPS method demonstrated high accuracy in vehicle state estimation.
• Performance was validated using real-world data from on-board vehicle sensors.
• Statistical analysis confirmed significant improvements over existing filtering and smoothing methods.

Conclusions:

• The proposed nGDPS offers a superior approach to VSE under non-Gaussian noise conditions.
• The method effectively handles complex noise distributions, leading to enhanced state estimation accuracy.
• nGDPS represents a significant advancement in smoothing techniques for automotive applications.