How to check if F/T sensors readings are correct

[KevinAStein]

Hi,
we are wondering how to check if the readings of the F/T from the analog:o are correct?
For example putting the robot in upright standing position on the ground, we see that the readings from right_foot\analog:o and left_foot\analog:o are pretty different, while we thought they should be pretty similar.

Thanks!
cc @hu-yue

[traversaro]

Hi @KevinAStein , consider that the measurements of the F/T sensors (as streamed by the /icub/***/analog:o ports ) are affected by an unknown offset. This offset is computed at the startup (and when the calib*** rpc function are called) by the wholeBodyDynamics and wholeBodyDynamicsTree modules.

To get some ports streaming measurements of the F/T sensors with this offset removed, launch the wholeBodyDynamicsTree module with the output_clean_ft option (you can simply add --output_clean_ft in the wholeBodyDynamicsTree option box in yarpmanager).

Once you get the measurement without the offsets, you can try to load the robot with some additional object with a known weight to check the measurement of the F/T sensors (the use of an additional weight is necessary because the known weight of the robot is already used by wholeBodyDynamics*** to estimate the F/T offset).

Consider that for two feet balancing scenario, this offset is computed when you launch the twoFeedStandingIdleAndCalib.sh script, as described in https://github.com/robotology/codyco-modules/tree/master/src/modules/torqueBalancing .

Furthermore, consider that apparently the offset computed for the F/T sensors in the lower limbs seems to be highly dependent on the load on the legs (i.e. if the robot is in single support on the left/right leg, in double support or with the root_link fixed to the pole). We are currently in the process of investigating this phenomenon with @fjandrad , for more info on this please check robotology-legacy/codyco-modules#160 .

[KevinAStein]

@traversaro thanks, but after launching the WBD3 with output_clean_ft how do we read the data without offsets? Is there a port from WBD3 where to read them?

[KevinAStein]

@traversaro just found it, sorry, was not seeing it because of a typo in the parameters.

[KevinAStein]

@traversaro if in zero upright standing position my filtered readings are
right_foot
2.409474 0.740165 -121.195665 -0.276714 0.25103 0.133824
left_foot
-2.026843 1.423318 -129.524288 -0.42297 0.26009 -0.047352
Would this be reasonable? The difference on the z axis is almost 10, which seems quite big, but maybe it's not?

[traversaro]

I think that are reasonable values. At least they are similar to measurements that we get on working iCub when we do double balancing demos (cc @DanielePucci for confirmation).

For the sake of completeness, consider that FT sensors are always affect by termal drift, so if you got the measurement some time after the calibration, this could also affect how much the sensor output is reliable.

[hu-yue]

Ok thanks, so 10 is not a big gap I assume.
We took the measurements right after the calibration, so I hope there was still not much thermal drift?
We also tried to load some weights on the robot handles and the measurement seems reasonable so I guess the sensors are fine. I wanted to check them because the torqueBalancing is behaving somehow differently, but this is another issue I'll try to take care of.

I'll close this.

[DanielePucci]

The difference in the readings looks reasonable to me.