Quaternion/Position order is not coherent between logger and thrift

[lrapetti]

Describe the bug
With @Zweisteine96 we noticed difference in the linear/angular serialization used used for iFeel data:

• in the thrift message we use angular/linear for the pose (see here and here), while for the velocity in the thift we use linear/angular (see here)
• In the sensors interface implementation the order seems to be again orientation/position (see here and here)
• In the logger, data are saved as position/orientation (see here and here)

The difference between logger and thrift has created some confusion.

This opens ups to some solutions:

• Document property the way data are logged and transmitted
• Change the order in the thrift (this will break backcompatibility of the dataset, even if it would be clear the order used)
• Change the order in the logger (same as previous)
• ... others?

Any suggestion/opinion @RiccardoGrieco @traversaro @diegoferigo ?

[traversaro]

If we want to change for harmonization, I would stick to linear/angular serialization as much as possible. Recently I also investigated integrating iDynTree and pinocchio in https://github.com/ami-iit/idynfor, and also in pinocchio the serialization used is linear/angular, see https://github.com/ami-iit/idynfor/blob/97ff77e6f148cabe7d95f51ee1c67958e9c9f918/doc/theory_background.md#rigid-body-velocity .

[lrapetti]

If we want to change for harmonization, I would stick to linear/angular serialization as much as possible

Yess, in general I agree with this. It would actually be a very quick change in the code.

The main issue, however, is that it would break all the datasets that we logged in the past using the yarpdatdumper which is the default way to store datasets. I don't know if there is a way to have a smooth transition.

[traversaro]

The main issue, however, is that it would break all the datasets that we logged in the past using the yarpdatdumper which is the default way to store datasets. I don't know if there is a way to have a smooth transition.

If we have them saved all somewhere, we could write a simple tool to convert the datasets between the old and the new version. But I do not know if doing that is worth the time/effort necessary to do that.

[lrapetti]

The main issue, however, is that it would break all the datasets that we logged in the past using the yarpdatdumper which is the default way to store datasets. I don't know if there is a way to have a smooth transition.

If we have them saved all somewhere, we could write a simple tool to convert the datasets between the old and the new version. But I do not know if doing that is worth the time/effort necessary to do that.

This solution is certainly feasible, but we have multiple datasets around so it might be difficult to track all the dataset changes