SLAM Algorithm Based on Variational Bayesian Noise Adaptive Kalman

Traditional filtering methods can achieve accurate results when the observations follow the assumption of a Gaussian distribution. However, in real-world motion scenarios, the shaking of the mobile robot platform and environmental disturbances often lead to outliers in the observations, resulting in heavy-tailed noise that deviates from the Gaussian distribution. In such cases, the performance of traditional filtering Simultaneous Localization and Mapping (SLAM) algorithms is significantly compromised due to the violated assumption. To address the presence of outliers in observations that do not conform to the Gaussian distribution, this paper proposes an adaptive SLAM algorithm based on the Student's t-distribution. Compared to the Gaussian distribution, the Student's t-distribution has thicker tails and is less sensitive to outliers, making it a more appropriate choice for modeling observations. By modeling the observations using the Student's t-distribution and employing variational Bayesian methods, the algorithm estimates the posterior probability distribution of the joint distribution of the mobile robot's state and observation parameters. This adaptation enables tracking of changes in observations and improves the accuracy of the algorithm. Simulation results demonstrate that when outliers are present in the observations, the VB-GCKF-SLAM algorithm achieves significantly higher localization accuracy compared to the Covariance Filtering SLAM algorithm (CFK-SLAM).