no-odom no-imu version

[kevinlisun]

By any chance, could you provide an example without the usage of odometry and imu? That is only use the 3D Lidar scans for localization.

Thanks!

[at-wat]

I guess it is not feasible with the current version. Expanding particle state from 6-dof pose (current impl.) to 12-dof pose+twist might help to realize it.

[kevinlisun]

Thanks! Is there a possibility to use vehicle odometry + Lidar without imu? We have some experimental data that imu is not available....

On 3 May 2018 at 07:52, Atsushi Watanabe [email protected] wrote:

I guess it is not feasible with the current version. Expanding particle state from 6-dof pose (current impl.) to 12-dof pose+twist might help to realize it.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/at-wat/mcl_3dl/issues/144#issuecomment-386197361, or mute the thread https://github.com/notifications/unsubscribe-auth/AF3LQNbQ2q1R464v0fQGhQ2DpBObhOpWks5tupsCgaJpZM4TtbDm .

-- Best Regards,

Li Sun (Kevin)

[at-wat]

I think LIDAR with odometry would be ok.

In this node, only acceleration part of the imu data is used for roll and pitch pose estimation. So, passing acceleration of (0.0, 0.0, 9.8) of dummy imu data would work assuming that roll and pitch are not very large for the car.

Note that if you are using Velodyne data, the parameter accum_cloud should be (number of pointcloud messages in one rotation) x N. Some more parameters should be modified: e.g. enlarge clip_*, downsample_*, map_downsample_*, odom_err_*. Also, odom_err_integ_*_sigma should be very large if the odometry is not accurate.

Currently, there is no document for these a lot of parameters. See: https://github.com/at-wat/mcl_3dl/blob/ccaf2451deead67907d8dd20815c98d791d5cf1b/src/mcl_3dl.cpp#L1472-L1623

[kevinlisun]

Thank you indeed for your help! I will give it a try :)

On 3 May 2018 at 11:05, Atsushi Watanabe [email protected] wrote:

I think LIDAR with odometry would be ok.

In this node, only acceleration part of the imu data is used for roll and pitch pose estimation. So, passing acceleration of (0.0, 0.0, 9.8) of dummy imu data would work assuming that roll and pitch are not very large for the car.

Note that if you are using Velodyne data, the parameter accum_cloud should be (number of pointcloud messages in one rotation) x N. Some more parameters should be modified: e.g. enlarge clip_, downsample_, map_downsample_, odom_err_. Also, odom_err_integ_*_sigma should be very large if the odometry is not accurate.

Currently, there is no document for these a lot of parameters. See: https://github.com/at-wat/mcl_3dl/blob/ccaf2451deead67907d8dd20815c98 d791d5cf1b/src/mcl_3dl.cpp#L1472-L1623

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/at-wat/mcl_3dl/issues/144#issuecomment-386233375, or mute the thread https://github.com/notifications/unsubscribe-auth/AF3LQAZ77qIibXwIoDqcXGyqZnsrFrkmks5tushwgaJpZM4TtbDm .

-- Best Regards,

Li Sun (Kevin)

[ielhmps7796]

@sunliamm hello,could you tell me the result of your experimert? I want to localize without IMU and Odometry too. Thanks~

[Oyssster]

hello, I have tested the version without odom, which means set without_odom:=true. It comes the errors that there is no tf from map to so many links. Is this code can run without odom?Thanks~