Simultaneous Localization and Mapping with Gaussian Technique

Abstract

Kalman filter (KF) is the estimator that minimizes the mean square error when the state and estimation dynamics are linear in nature, given the process and measurement noise are demonstrated as white Gaussian. In the real world, it is always the case that either the process or measurement noise is nonlinear. Extended Kalman Filters are proposed for such scenarios, but they are only suboptimal solutions as they linearize the nonlinear data using methods like Taylor series expansion. In any case, in reality, one experiences an enormous number of situations where either the procedure or estimation model (or both) is nonlinear. Using recursive Gaussian inference, identifying noisy features and dynamically calculating the relative positions between them can be done. By applying more variations to the prior distribution calculations method, noisy sensor data can also be processed into small local maps and formed into a consistent global map. The Kalman Filter-based or pure Recursive Bayesian approach-based mapping algorithms are capable of extracting very few features from the sensor data which are not sufficient for effective simultaneous localization and mapping in noisy and multistoried environments. Every point in the feature is represented as a vector with coordinates and orientation which is given to the Bayesian inference method for predicting the position of the robot when the inputs given are noisy sensor data and noise-level limit.