some trivial questions about direct_lidar_odometry HOT 4 CLOSED

Hi @narutojxl --

• The difference quaternion between two quaternions can be computed by multiplying one with the inverse of the other.
• You are right; thanks for the catch! We've fixed this in 1.3.1
• Good catch again -- the update should come before the covariance calculation. Fixed in 1.3.1
• Since quaternions are a double map of SO3, a quaternion q and -q represent the same rotation so sign flips ultimately won't affect the solution. Sign flips are not preferable for controllers, however, so we chose to keep the check prior to publishing to minimize number of operations in the main loop.

Thanks again very much for your questions and bug catches.

from direct_lidar_odometry.

Thanks author fast reply.
For the first question in code, Eigen::Quaternionf dq = this->rotq * (closest_pose_r.inverse());, does this dq have a actual physical meaning? Because this->rotq.inverse() * closest_pose_r represents curr scan to cloest keyframe rot, closest_pose_r.inverse() * this->rotq represents cloest keyframe to curr scan rot, they all have a actual physical meaning.

from direct_lidar_odometry.

Yes, dq is the relative rotation which brings closest_pose_r to this->rotq, such that

dq * closest_pose_r = this->rotq

or

dq = this->rotq * closest_pose_r.inverse()

Since we're interested in the magnitude between the current pose and a keyframe, it doesn't matter which "direction" we go since we take the absolute value of this angle (lines 1145, 1149, 1151).

from direct_lidar_odometry.

OK, I see its meaning is a global delta_rot which transforms cloest_pose_r into curr rot. Thanks very much! My problem is solved, close it now.

from direct_lidar_odometry.