Doubts on pose_sub usage and meaning

Technical Questions URScript
1

Hello,

I’m trying to understand how to use pose_sub together with speedL and trying to imitate the pose_sub functionality in Python.

I’ve written a small program which has 4 variables containing 4 poses.

var_1 is the base pose, from where all the movements start.
var_2 represents a position change from var_1.
var_3 represents an orientation change from var_1
var_4 represents a pose change from var_1

First I do te movements using MoveL commands, and then I try to reproduce them with speedL commands.

It seems odd to me that in order to imitate the MoveL commands and achieve the same final poses with pose_sub and speedL, rotations (coming from pose_sub) have to change the sign while translations (coming from pose_sub) keep the same sign.

Why is this?

Besides, a second question is, how is the orientation subtraction computed?

I’ve tried this using Python:

  1. From var_1 and var_3, convert Rx, Ry and Rz from axis angle representation to quaternions (also tried using roll, pitch and yaw).
  2. In quaternion form, compute de rotation difference.
  3. Take this orientation different back to an axes angle representation.

pose_sub_test
pose_sub_test1583×771 74.1 KB

2

Hello,

I found the solution to the second question: how is the orientation subtraction computed so that it can be used to feed an speedL command?

In pseudo-code, it’d be like this:

  1. From var_1 and var_3, convert Rx, Ry and Rz from axis angle representation to quaternions (q1 and q3)
  2. In quaternion form, compute de rotation difference as: q_diff_orientation= quternion_multiplication(quaternion_inverse(q3), q1)
  3. Take this q_diff_orientation back to an axes angle representation to feed speedL.

I still find intriguing why the difference of sign when using translations or rotations.

Best,

3

Hi @dalvarez
pose_sub is basically created so the following expression becomes true

pose_add(pose_sub(p_1, p_2), p_2) == p_1

What is your use case for the speedl functionality?

4

Hello Ebbe,

Thanks for the explanation.

The high level task I’m trying to accomplish is:

  • I record the X,Y,Z,rx,ry,rz information of a trajectory performed using RTDE.
  • Then, I want to resproduce it. While repoducing, there’ll be small differences in the path that will appear due to the movement of the work piece (this is why I cannot record joint positions). I detect the need to slightly change the path using the force sensor.
  • This reproduction is also controlled using RTDE.

So, first, I though I would perform a control loop in my computer like:

  • Read actual_pose
  • Compute goal_pose-actual_pose (this is why I wanted to learn to reproduce pose_sub in python)
  • And send speedL commands

The robot controller would basically read the inconming information and execute speedL commands.

I know I could also try a similar control loop using servoj(get_inverse(goal_pose-actual_pose)), adjusting each loop how much I want to advance.

I’m trying to figure out which one fits better in my applitcation.
I started with speedL because I find it more intuitive.

Best,

5

pose_sub is implemented as the inverse (or complement to) pose_add.

The translation part is simply subtracted, e.g. T3 = T1 - T2. The rotation part is converted to a matrix (R1) and the second rotation part is converted to a matrix and then inverted, finally the rotation matrices are matrix multiplied (R1xR2inv) to get R3. The whole is then converted back to a pose.