Transactions on Machine Intelligence 2821-1693 3 2 2020 06 01 Design and Implementation of a Discrete-Time Unscented Kalman Filter (DTUKF) Based on Genetic Algorithm to Enhance the Performance of Nonlinear Navigation Systems 64 80 206186 10.47176/TMI.2020.64 EN A. Talebi Sheikh Sarmast Control Department, Faculty of Electrical Engineering, Shahid Beheshti University, Tehran, Iran N. Samadian Electronics Department, Faculty of Electrical Engineering, Semnan University, Semnan, Iran H. Zeinali Control Department, Faculty of Electrical Engineering, Shahid Beheshti University, Tehran, Iran S. M. Mousavi Mirkalaei Professor, Electronics Department, Faculty of Electrical Engineering, Iran University of Science and Technology, Tehran, Iran Journal Article 2020 02 10 Nonlinear inertial navigation systems are considered among the most important navigation systems, featuring advantages such as independence from external communication, real-time speed and position calculation, and suitable bandwidth, making them very popular in vehicle navigation. These systems consist of an inertial measurement unit comprising three orthogonal accelerometers and three gyroscopes used to determine the position, speed, and direction of vehicle movement, calculating the vehicle''s position with minimal error over short distances. However, over time, due to errors in the gyroscopes and accelerometers and consecutive integrations of their outputs, the estimated position error increases. To achieve higher accuracy, especially in long-term navigation, a global positioning system (GPS) is used for its complementary properties with the inertial navigation system (INS). This paper discusses and analyzes the application of two Kalman filters linear Kalman filter and unscented Kalman filter on nonlinear navigation systems. A discrete-time unscented Kalman filter (DTUKF) is utilized to estimate the position, speed, and state of a nonlinear system, which in this case is an integrated navigation system. By accurately selecting a set of sigma points from a Gaussian distribution and propagating these points through the nonlinear function, estimation is performed with significantly higher accuracy. The unscented transformation allows for the selection of the distribution of these points and control of higher-order error using design parameters. Comparing this filter with the linear Kalman filter reveals the superior performance of the unscented Kalman filter due to not linearizing the nonlinear system, thereby reducing system error. Notably, this paper is the first to use a genetic algorithm to optimize the Q and R noise matrices to achieve minimal variance and optimal mean convergence to reference values. https://www.tmachineintelligence.ir/article_206186_1aa7d6a7eee11d62406673b1eb20c6eb.pdf