0
$\begingroup$

I want to implement the J+RRT algorithm for which I need to use the Twist command to navigate the robot in Euclidean space.

First I calculate the difference in rotation matrices from the end-effector frame T1 to the goal frame T2 like so:

Tdiff * T1 = T2  // * T1^-1
Tdiff * T1 * T1^-1 = T2 * T1^-1
Tdiff = T2 * T1^-1

I am using the Frame separated as a rotation and translation. Effectively this means that

T1^-1 = [R1^T -t1; 0 0 0 1]

RPY seems to be given in extrinsic x, y, z angles, which means that I need to convert it to intrinsic angles for the Twist command. This means that the RPY angles that I get from the Tdiff I would need to convert to my R1 rotation frame. Is my thinking correct?

The code I'm using is this:

KDL::Rotation inv_src_rot = src.M.Inverse();
double x, y, z;
KDL::Rotation diff_rot = tgt.M * inv_src_rot;
diff_rot.GetRPY(x, y, z);
twist.rot = src.M * KDL::Vector(x, y, z);

This is the code for the RPY calculation: https://github.com/orocos/orocos_kinematics_dynamics/blob/master/orocos_kdl/src/frames.cpp#L237-L260

EDIT: Right as I posted this I had another idea, which seems to me more correct.

If my goal is to go towards the goal rotation tgt.M in the end-effector reference frame src.M, it would make sense to calculate the RPY angles from the rotation

src.M * tgt.M,

since in this case the extrinsic frame has now become src.M. Now when I calculate RPY it will be the rotation tgt.M given in the src.M frame, if I'm not mistaken?

KDL::Rotation diff_rot = src.M * tgt.M;
diff_rot.GetRPY(x, y, z);
twist.rot = KDL::Vector(x, y, z);
$\endgroup$

2 Answers 2

0
$\begingroup$

Angular velocities are strongly related to angle-axis description of a 3D rotation.

Instead of using RPY, just use the angle-axis vector (as in angle times unit axis) of the rotation error (R1.R2^T). Then the angular velocity to drive this vector to 0 is just proportional to the vector (e.g. this is a proportional velocity control in SO(3)).

$\endgroup$
1
$\begingroup$

Your description is a bit confusing and I am unable to post a comment for clarification based on my reputation score, so I will try my best to help.

From your code, it looks like you only want to rotate the end-effector and not translate (is this true?).

Given that your end-effector is at orientation src.M in the world frame, and wants to go to a new orientation tgt.M in the end-effector frame, the new orientation in the world frame will be src.M * tgt.M, which is what you mention in your edit. But then, these will give you RPY angles, whereas Twist concerns with angular velocities. You need to implement a control that would vary your angular velocities (Twist) to get you to the required angle.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.