Torque readings from DX200 are not continuous.

[nexus1203]

After following the instructions on this comment #174, I collected the torque data using two method, first the controller torque method another is the Mregister method. However both the methods return torque readings with a significant delay. However, Joint readings are still continuous. I am hoping devs can shed some light on this, if this is a controller limitation? I have ordered MotoPlus from the local dealership. I am hoping that may solve the issue. However, the comparisons are attached as follow.
The joint angles are in pulse. sorry for the label.
DX200_torque.csv

DX200_Mregister.csv

[ted-miller]

Hi @nexus1203
I think what you're seeing is caused by the frequency that feedback data is published from motoros to motoman_driver.

motoman/motoman_driver/MotoPlus/StateServer.h

Line 35 in 7860ff5

#define STATE_UPDATE_MIN_PERIOD 25 // Time delay between each state update

I have ordered MotoPlus from the local dealership

With the SDK, you can decrease the STATE_UPDATE_MIN_PERIOD and you should see immediate improvement.

[gavanderhoorn]

I don't understand the plot(s).

Is the X axis time in ms, or is it just counting periods (which are 20 ms apparently)?

Additionally, what does this mean exactly:

After following the instructions on this comment #174, I collected the torque data using two method, first the controller torque method another is the Mregister method.

how are you getting the data from the controller?

Is this still MotoROS with Simple Message, or something else?

[ted-miller]

you can observe that controller is updating the position data continuously, however torque data stays the same for several req/res cycle

To be honest, I didn't think this was true. But I ran a quick test and you are absolutely correct. This is seems to be a limitation of the M-Register method.

(Built-in pendant oscilloscope. Horizontal resolution: 100 ms / div)

I have seem similar behavior with other API's. But, I'm surprised that the feedback-position behavior is so different from the feedback-torque behavior when using the same method of data collection (m-registers).

As per my understanding of servo motor controllers, yaskawa sigma5 drivers (the one DX200 uses) are capable of updating torque readings every 1ms

However, I suspect if the controller doesn't update the torque readings then how would I be able to get any better resolution even at higher frequency?

As you can see from the blue line in the chart, the controller is receiving torque data at a much higher rate than what is published to the m-registers. We just need to find the best API for pulling that data out of the controller.

I'm pretty sure that the MotoPlus API, mpSvsGetVelTrqFb updates at much higher rate (I think 500Hz). But, I'm not currently setup to test/validate that statement. I should be able to run a test on Wednesday or Thursday.

[nexus1203]

As you can see from the blue line in the chart, the controller is receiving torque data at a much higher rate than what is published to the m-registers. We just need to find the best API for pulling that data out of the controller.

Yes, certainly the controller is updating at much higher rate. This is promising.

I'm pretty sure that the MotoPlus API, mpSvsGetVelTrqFb updates at much higher rate (I think 500Hz). But, I'm not currently setup to test/validate that statement. I should be able to run a test on Wednesday or Thursday.

Thank you again for your help. This will give me a head start on planning my data collection strategy while I wait for the MotoPlus SDK.

[gavanderhoorn]

I would suggest to take MotoROS and update Ros_SimpleMsg_JointFeedback(..) to call Ros_CtrlGroup_GetTorque(..) instead of Ros_CtrlGroup_GetFBServoSpeed(..).

That would replace the velocity feedback in JOINT_FEEDBACK_EX messages with the joint torques.

At that point, use ra-mtp-ntnu/moto to start a regular StateServer connection, and you'd receive joint torques at 40 Hz. If you update the constant @ted-miller mentions in #508 (comment), you could go even higher.

It wouldn't be the nicest solution (as you'd be putting torques in the velocities field), but I don't believe anything else would be needed.

[ted-miller]

I created a very simple MotoPlus application that reads the torque every 2 ms and writes it to an M-Register (just so that I can see it on the scope).

void mpTask1(void)
{
    MP_GRP_AXES_T dst_vel;
    MP_TRQ_CTL_VAL dst_trq;
    LONG status = 0;
    int i;
    MP_IO_DATA regData;

    regData.ulAddr = 1000000 + 40;

    while (TRUE)
    {
        mpTaskDelay(2 / mpGetRtc());

        bzero(dst_trq.data, sizeof(MP_TRQCTL_DATA));
        dst_trq.unit = TRQ_NEWTON_METER; //request data in Nm

        bzero(&dst_vel, sizeof(MP_GRP_AXES_T));

        mpSvsGetVelTrqFb(dst_vel, &dst_trq);

        regData.ulValue = ((double)dst_trq.data[0][0] * 0.001);

        mpWriteIO(&regData, 1);
    }
}

It seems that mpSvsGetVelTrqFb is updating at the rate of the robot's interpolation clock. For a single robot arm, the default interpolation clock is 250 Hz. (That will get slower if additional axes are added to the controller.)

This is the best we can do with the API's that are available.

[nexus1203]

For a single robot arm, the default interpolation clock is 250 Hz.

@ted-miller Thank you again. I think 250Hz seems reasonable as the IO output rate is also 4ms according the DX200 manual. Also, thanks for the sample code, I think this will help me a lot once I get the SDK.

[gavanderhoorn]

Looking at the data, I find it suspicious it only shows whole numbers for the torque data.

I don't remember that being the case for torques reported by MotoROS / the other project.

@ted-miller?

[gavanderhoorn]

But, then the data plotted here is in micro-Nm?

I would not expect that.

[gavanderhoorn]

Originally by @ted-miller:

BTW. The current version of MotoROS is not using the M-Registers to get torque data. It is using a MotoPlus API called mpSvsGetVelTrqFb.

[ted-miller]

But, then the data plotted here is in micro-Nm?

No. The M-register method outputs a percentage of rated-max torque.
Ex: 5 = 5 %

There is no way (that I'm aware of) to get Nm into the registers.

[gavanderhoorn]

And the HSES function? Because that .csv seems to contain similar numbers.

[ted-miller]

Most likely. It is a common unit for other torque API's.

The HSES doc does not specify: