ywl0720 / i2ekf-lo Goto Github PK

Hi,

Thanks for the code. I am looking into the maths of the prediction step, which involves calculating the Jacobian F_x and F_w.

I found your calculation of F_x with respect to the rotation and angular velocity errors here:

Do you have the technical note on these calculations? Just to share some results.

dx_tilde_{k+1} / d_rotation_error = Exp(omega_hat, dt)^{-1},

which is equivalent to Exp(omega_hat, -dt)

My calculation of dx_tilde_{k+1} / d_angular_velocity_error shows that it should be Jr(angular_velocity_estimate * dt)dt, which can be approximated to Eye(3) * dt as in your code.

Thank you for your great work!

I use the "mid360.yaml" in your code and set "adaptive_cov_use" true. When tested with "dance.bag" , the odometry drift quickly and cannot reproduce the mapping result in Fig. 6d of your paper.

Could you tell me if "mid360.yaml" is the right config file for "dance.bag"?

感谢您在 I2EKF 和 LiDAR 里程计上的工作。我注意到缩写“I2EKF”可能会导致误解，容易让人联想到 ICP 和迭代卡尔曼滤波更新阶段的优化。
建议考虑使用更清晰的缩写，如 RIEKF（Recursive Iteration Extended Kalman Filter），以突出迭代的递归性质。这可以帮助避免混淆。