search

hku-mars / FAST_LIO

A computationally efficient and robust LiDAR-inertial odometry (LIO) package
GNU General Public License v2.0
2.8k stars 940 forks source link

Question about discrete model in fast lio #36

Closed LeisureLei closed 3 years ago

LeisureLei commented 3 years ago

Thanks for your great job! I test fast lio2 on KITTI dataset and get nice results. But I have a question about discrete model when I read fast lio paper. image According to formula (2) (3), we can get P{t+1} = P{t} + V{t} * δt , why not P{t+1} = P{t} + V{t} δt + 1/2 a_{t} * δt^{2}?

XW-HKU commented 3 years ago

Thanks for your interests. You are right. Fastlio1 uses the zero order discretion，Fastlio2 uses the first order model. But they are really very similar since the imu sample time is always very tiny.

LeisureLei commented 3 years ago

Thanks for your interests. You are right. Fastlio1 uses the zero order discretion，Fastlio2 uses the first order model. But they are really very similar since the imu sample time is always very tiny.

Thanks! And I have another question about the computation of mattix F_x and F_w in appendix A. Are there some materials describing specific process?

XW-HKU commented 3 years ago

Thanks for your interests. You are right. Fastlio1 uses the zero order discretion，Fastlio2 uses the first order model. But they are really very similar since the imu sample time is always very tiny.

Thanks! And I have another question about the computation of mattix F_x and F_w in appendix A. Are there some materials describing specific process?

I think the computation of the F_x and F_w would be very easy, if you regard appendix A as a simple chain-rule of partial differential. And IKFOM's appendix A&B has a simple example for the computation of chain-rule of partial differential combining the boxplus operation, may help you.

LeisureLei commented 3 years ago

Thanks for your interests. You are right. Fastlio1 uses the zero order discretion，Fastlio2 uses the first order model. But they are really very similar since the imu sample time is always very tiny.

Thanks! And I have another question about the computation of mattix F_x and F_w in appendix A. Are there some materials describing specific process?

I think the computation of the F_x and F_w would be very easy, if you regard appendix A as a simple chain-rule of partial differential. And IKFOM's appendix A&B has a simple example for the computation of chain-rule of partial differential combining the boxplus operation, may help you.

Thanks!

LeisureLei commented 3 years ago

@XW-HKU hi, I wonder what the difference about matrix F_x and F_w between fast lio and Quaternion kinematics for the error-state KF. I try to compare them. And some parts sre same, some parts are different.

XW-HKU commented 3 years ago

@XW-HKU hi, I wonder what the difference about matrix F_x and F_w between fast lio and Quaternion kinematics for the error-state KF. I try to compare them. And some parts sre same, some parts are different.

The main differences i think is using rotation matrix or quaternion.

LeisureLei commented 3 years ago

@XW-HKU hi, I wonder what the difference about matrix F_x and F_w between fast lio and Quaternion kinematics for the error-state KF. I try to compare them. And some parts sre same, some parts are different.

The main differences i think is using rotation matrix or quaternion.

Got it!

HaisenbergPeng commented 2 years ago

@XW-HKU hi, I wonder what the difference about matrix F_x and F_w between fast lio and Quaternion kinematics for the error-state KF. I try to compare them. And some parts sre same, some parts are different.

The main differences i think is using rotation matrix or quaternion.

Hi, Mr. Xu, there is a very good blog about ESKF deduction based on SO3, which is written by Dr. Gao, Xiang. Based on his deduction, there should be no "-A(w_ideltaT)^TdeltaT" for equation 7 in your 2021 RAL fastlio paper. image What do you think?

ysingfun commented 1 year ago

@XW-HKU hi, I wonder what the difference about matrix F_x and F_w between fast lio and Quaternion kinematics for the error-state KF. I try to compare them. And some parts sre same, some parts are different.

The main differences i think is using rotation matrix or quaternion.

Hi, Mr. Xu, there is a very good blog about ESKF deduction based on SO3, which is written by Dr. Gao, Xiang. Based on his deduction, there should be no "-A(w_i_deltaT)^T_deltaT" for equation 7 in your 2021 RAL fastlio paper. image What do you think?

I'm also curious that it's not in the code, but it's in the paper.Again, i should be clear about the formula.

zhouqiuhua commented 2 months ago

Thanks for your great job! I test fast lio2 on KITTI dataset and get nice results. But I have a question about discrete model when I read fast lio paper. image According to formula (2) (3), we can get P{t+1} = P{t} + V{t} * δt , why not P{t+1} = P{t} + V{t} δt + 1/2 a_{t} * δt^{2}? Hello boss, may I ask which sequences from the Kitti dataset you tested on Fast-LiO2? I tested all the sequences, but only 05, 06, 07, and 09 were successful. The others were all quite outrageous

zhouqiuhua commented 2 months ago

Hello boss, may I ask which sequences from the Kitti dataset you tested on Fast-LiO2? I tested all the sequences, but only 05, 06, 07, and 09 were successful. The others were all quite outrageous