Looking for a reference use of PoseWithCovariance: any pkg using it?

asked 2012-03-30 23:45:40 -0600

Jose Luis Blanco

288 ●8 ●14 ●16 http://www.ual.es/~jlblanco/

updated 2014-01-28 17:11:35 -0600

ngrennan
1 ●1 ●1

Hi all,

PoseWithCovariance has a 6x6 covariance matrix which, according to REP 103, corresponds to these variables in this order: (x, y, z, rotation about X axis, rotation about Y axis, rotation about Z axis). With rotation around fixed axes, not like in the "common" yaw/pitch/roll various conventions.

I'm working in a pkg for covariance transformations (pose_cov_ops) so it would be great if I could compare my numerical results with any other existing ROS packages that output PoseWithCovariance at present. Checks would go in the package unit tests.

More explicitly: I'm looking for some "stable" ROS package which computes covariances in 6D, so I can peek its source code and make sure I'm following the right convention.

Any recommendation?

If there're none, I'll assume the 3x3 rotation matrix is as follows:

~~R = R_x(roll) * R_y(pitch) * R_z(yaw)~~

R = R_z(yaw) * R_y(pitch) * R_x(roll)

where matrices are in the ~~inverse~~ same order than the non-global axis, yaw-pitch-roll convention.

PS: wouldn't it have made more sense to keep a 7x7 covariance for the unit quaternion form? In general, equations for uncertainty propagation in this form are far simpler than for Euler angles. Is there room for a PoseWith7x7CovarianceStamped yet?

edit retag flag offensive close merge delete

Comments

Just for the records: the numerical values of yaw/pitch/roll in the dynamic-axes (rotating) convention are exactly the same than those in the roll/pitch/yaw fixed-axes convention. So a conversion of covariance matrices between both formats becomes just a permutation. Hope it may help someone else.

Jose Luis Blanco ( 2012-04-11 15:02:15 -0600 )edit

add a comment

1

answered 2012-06-25 06:50:31 -0600

Pablo Iñigo Blasco

2982 ●74 ●121 ●128 http://www.linkedin.co...

updated 2012-07-03 10:23:31 -0600

AFAIK there is no ROS package which make the transform between two poses with uncertainty (perhaps http://ros.org/wiki/bfl but I don't think so). Because of this the package pose_cov_ops seems unique. I belive that the MRPT ([[mrpt-ros-pkg]]) in which the pose_cov_ops package is based looks very powerful handling poses with uncertainty and ROS should take advantage of this fact.

I think that this package should be integrated in the generic http://ros.org/wiki/tf2 package as a plugin in a similar way that the kdl package was integrated (see https://kforge.ros.org/geometry/experimental/file/40be50217595/tf2_geometry_msgs/include/tf2_geometry_msgs/tf2_geometry_msgs.h). Perhaps the tf2 package should be refactored to support more explicitly the inverse tranformation or at least the unary inverse transform.

Edit: Definitely I think you can find what you are looking for in the source code of the package http://ros.org/wiki/robot_pose_ekf

Regards.

edit flag offensive delete link

add a comment

1

answered 2012-06-25 06:59:17 -0600

dornhege

31395 ●134 ●284 ●498

updated 2012-06-25 06:59:29 -0600

There is tf_uncertainty that might do what you are looking for.

edit flag offensive delete link

Comments

I don't locate in the package source code the gaussian error propagation (using Jacobians) produced by the nonlinear tranforms. The resulting uncertainty after a transform is represented by a pose cloud(particles) and not as a PoseWithCovariance. In such case it is not what José Luis is looking for.