search

robotology-legacy / codyco-modules

Whole-body Compliant Dynamical Contacts in Cognitive Humanoids
www.codyco.eu
Other
19 stars 13 forks source link

[WholeBodyDynamicsTree] Inconsistent measurement of joint friction #82

Closed rbonghi closed 7 years ago

rbonghi commented 9 years ago

I am using Whole Body Dynamics (WBD) and Whole Body Dynamics Tree (WBDT) to measure friction on joint (test on iCubGenova03 and iCubGenova01), but I have a problem to acquire data.

If I acquire torque and velocity in different position from same joint I measure a drift in friction, because I expect the same result for any position.

For these experiments I calibrated (WBD or WBDT) in the fixed position, I put the joint in idle and finally I moved manually near the the position calibrated.

I have attached in this issue a plot with this measures. These experiments were carried out on iCubGenova03 (legs V1) on left leg hip pitch and I used WholeBodyDynamics (WBD) and in a second experiment I used WholeBodyDynamicsTree (WBDT).

iCubGenova03 Whole Body Dynamics icubgenova03-wbd iCubGenova03 Whole Body Dynamics Tree icubgenova03-wbdt

I measure a drift only in WBDT. Different result in iCubGenova01 (legs V2) because I measure a drift for both, WBD and WBDT.

iCubGenova01 Whole Body Dynamics icubgenova01-wbd

iCubGenova01 Whole Body Dynamics Tree icubgenova01-wbdt

I think the problem is in dynamic model on robot, because I would like to measure the same result for any position of joint.

DanielePucci commented 9 years ago

A word of clarification.

In view of the important differences in the angles at which you are taking the measurements, I would not claim that the "result should be the same" a priori. By the way, what do you mean by result?

The main point here is that at some configurations the measured friction doest not oppose the motion (e.g. Figure 2a, "iCubGenova03 wholeBodyDynamicsTree), and this clearly shows that there's a problem at the measurement level. Being these measurements based on the dynamic model, the problem may be in the model as well. Clearly, I do agree that if the measurements are taken by moving the joint really close to a specific angle, the function tau = tau (dot (q)) should be odd in the first approximation.

rbonghi commented 9 years ago

I repeat again this experiment on iCubGenova01 and I have a significant error when I put the robot in equilibrium with a one foot or the robot is in equilibrium in double support. Leg - Left - Hip - Roll onetwofeet

I have a different offset by 3 [Nm] in single suppport and -7 [Nm] in double support.

Why do I measure torques only positive or only negative friction?

DanielePucci commented 9 years ago

After the model for iCubGenova02 is finished, we can perform again the parameter identification, and see if this issue still arise. In the negative case, we can close it.

traversaro commented 8 years ago

@DanielePucci you had this problems with the new model?

DanielePucci commented 8 years ago

We haven't re-estimated the motor and friction parameters using the new model. Let's put this task on the to-do list.

traversaro commented 7 years ago

Probably it is/was a problem related to https://github.com/robotology/codyco-modules/issues/160 . Closing unless we have some action to perform related to this.