Sensor Fusion Study - Ch14. The Unscented Kalman Filter [Sooyoung Kim]
Download as PPTX, PDF0 likes313 views
This document provides an overview of the unscented Kalman filter (UKF), including its advantages over the extended Kalman filter (EKF). It discusses how the UKF works by using sigma points to capture the mean and covariance of estimates with nonlinear transformations more accurately than the EKF. It also provides examples of using the UKF for state estimation in localization problems with robot odometry and sensor fusion.
Sensor Fusion Study - Ch14. The Unscented Kalman Filter [Sooyoung Kim]
- 1. Unscented Kalman Filter Example Based
- 2. Content ● Simple Conception of UKF ● Example Based UKF ● Real World Example
- 4. EKF - Linearization, Jacobian
- 5. EKF의 단점 ● Linearized Mean/Cov != Nonlinear Mean/Cov ● Jacobian 계산이 너무 오래 걸림 - Stink!! ● 너무 복잡한 model, 미분할 수 없는 수식이 들어간 모델에 취약
- 6. Stink!
- 7. Unscented Kalman Filter and Sigma Points
- 8. ● How pick points ● How many points ● After then...
- 11. Example - CTRV Model
- 12. Generate Sigma Points - 2D
- 13. Generate Sigma Points - CTRV Model
- 14. Generate Sigma Points Effect of Design parameter
- 15. Various Form...
- 16. Take account process noise Kalman Filter에서는… http://campar.in.tum.de/Chair/KalmanFilter
- 17. Augmentation
- 18. Sigma Points with Augmentation
- 19. Predict Sigma Points Based on State Model http://campar.in.tum.de/Chair/KalmanFilter Example...
- 20. Predict Mean and Covariance
- 21. Recap
- 22. Predict Measurement 5 state => 7 * 15 matrix (augmented) => 5 * 15 matrix & new mean/cov (predict) => 3 * 15 (projection to measurement space) => calc mean & cov again
- 23. Update State
- 24. Process Noise…??? ● The process noise is usually treated as a tuning parameter to adjust the gain of the Kalman filter to smooth either more or less the data. ● want to respond quickly to changes. => Upper Process Noise
- 27. Odometry & Sensor Fusion https://github.com/udacity/robot_pose_ekf https://www.researchgate.net/figure/Paramet ● robot_pose_ekf => deprecated ● robot_localization
- 28. Q & A
- 29. Reference ● The UKF exposed: How it works, when it works and when it's better to sample ● Udacity - Sensor Fusion Nanodegree ● cra-ros-pkg/robot_localization
- 30. Thank you :)