Shaking iCub - update of config files

[Tobias-Fischer]

Hi all,

as reported in https://github.com/robotology/icub-support/issues/30, the config files of various robots have to be updated. Firstly, all robots should use TORQUE_CONTROL, and the old source for the other parameters should be deleted (or at least the bug should be corrected). Furthermore, for robots mounted on the iKart, the parameters Kp, Ki, and Kd need to be changed. It would be great if @jypuigbo could report the values which are used on iCubBarcelona01 now.

@pattacini, could you please assign someone for this issue?

Thanks, Tobi

[pattacini]

I've marked this as urgent. I think @randaz81 knows how to deal with it.

[super-ste]

In order not to change the PID paramenters (ki, kp, kd), you can lower the power supply voltage to a value closer to 38.5 -- 39 V and exclude the battery from the setup.

You will have to lower the PIDs if you plan to use the iKart autonomously and with the battery. In such case the voltage is set by the battery (>53 V when fully charged).

@randaz81 @pattacini @jypuigbo correct me if I'm wrong.

[Tobias-Fischer]

Hi @super-ste, as we are using the iKart with the battery every now and then, changing the power supply voltage is unfortunately not an option for us.

@jypuigbo could you please report the values you use for ki, kp and kd?

Thanks, Tobi

[pattacini]

@Tobias-Fischer the PID output is given in terms of machine ticks in the range [-max_output, max_output], where max_output (historically set equal to 1333 ticks) corresponds to the voltage full scale, i.e. 39 V or 53 V depending on the current power supply setting.

Therefore, multiplying the values of all your PID gains by a factor of 39/53 should do the job in a way that has at least a bit of rationale, whatever rules others might have followed to cut their gains. Hence, I would recommend you to start with this.

Further, I would also suggest to modify directly those values in the repository (now you can), to avoid future problems.

[Tobias-Fischer]

Hi,

I just pushed the modified configuration files for iCubLondon01 (see https://github.com/robotology/icub-main/commit/900b82e4484a3917218c091d3267311428f543a5). The robot seems to run fine with the new values.

However, I found that all values of ki and kd are set to 0, which makes the PID controller to a P controller, isn't it? Or are default values used if the values are 0?

I'll leave this ticket open until the TRQ_PIDS issue has been resolved.

Best, Tobi

[pattacini]

@Tobias-Fischer I'd say that also POS_PIDS should be affected by the gain reduction.

Apparently, when in torque mode there's only a P controller active, yep.

[Tobias-Fischer]

Hi Ugo @pattacini, thanks for pointing that out! I totally missed that these values need to be changed for the POS_PIDS as well, as I've only discovered the bug for TRQ_PIDS. I'll change the config files accordingly tomorrow.

[Tobias-Fischer]

Fixed in https://github.com/robotology/icub-main/commit/eaa9fb95d40a0c89e0d3a498b4b3edf06ab1a4f0. Thanks @pattacini!

[randaz81]

@pattacini This topic will be discussed in IFI Day 3.

[randaz81]

Hi @Tobias-Fischer, i'm going to revert your modifications to icub_all.xml. That file has a special meaning, it's not only parsed by robotInterface but also by a tool which updates the iit database. I am thus pushing to the repo the following solution: I created a new file icub_ikart.xml which disables the legs, and I modified the pointer contained in robotInterface.ini Btw, we encourage to use the pull request workflow for critical directories such as icub-main/app/robots. Thank you!

[pattacini]

:+1:

[Tobias-Fischer]

Hi @randaz81, Thanks for your changes. We've had this setup locally for a long, long time, and I didn't know about this tool you are talking about. We'll make sure to make our changes in icub_ikart.xml in the future.

Thanks, Tobias