what special need to modify the URDF file when identify other robots about dynamic_calibration HOT 4 CLOSED

Hi! My URDF parser is by no means perfect and general. However if the code passes all the tests, then URDF parser should not be an issue.
Also drive gain identification requires extra attention in terms of load selection and optimal trajectory design. First of all I would contact the manufacturer and ask for drive gain parameters. Next try to find those parameters in literature, and only then try to identify it myself.

from dynamic_calibration.

Hello! thank you for your answer. now I am trying to get the dynamic parameter on the ur16e robot according this paper.

I have four unclear questions:

1）I have asked the ur agent in china for drive gain parameters, but they don't provide the drive gain parameters , so I have to identify drive gains myself(I think they may be unprofessional,so they don't know these parameters) , or do I need to directly contact with UR office via email(I have no UR offical contact information or can you recommend some email of UR) ？

2）according to this paper , when I try to identify the driver gain parameters, I set the end load to 1.494kg instead of 2.805kg(in your paper) , will it have any bad effects, or is it too light(1.494kg)? Do I need to increase its mass(I am not sure about its mass is suitable or not now)?

3）I get the optimize trajectory (set N = 7 T = 25 & N = 14 T =50 & N = 20 T = 30 ) , and collect data in velocity control mode based on ros , filter the data and run estimate_drive_gains.m ,I get the drive gains like this: [10.0000 ,0.7198, 0.6910, -0.1488, -0.1916, 0.7170] , [9.999999999728000;0.645280136776565;0.546376701348147;-0.713932783414374;1.797815696467742;1.832032974481469] , it likes that some parameters are too small, and some are negative, I only did a few sets of experiments now（the robot has some malfunctions now）, I’m not sure if it’s caused by inappropriate settings of N and T (if true , I will collect more data later)? when i run dynamic parm identify script ,the relative residual error like this : [69.261201137938290;60.505411253048740;58.151022251522654;5.486685542589994e+02;2.701281946217017e+02;50.240006692999070] (the error is too big)
(also I want to ask the joint data sampling frequency whether influences the experimental result( I get data at 500Hz instead of
100Hz)).

4）about the urdf parser , first i run the scripts , I get the M , C , G , Y, Y_base matrix , and then run test_rb_inverse_dynamics(path_to_urdf) test_base_params(path_to_urdf) return Rigid Body Base Dynamics Test - OK! Rigid Body Inverse Dynamics Test - OK! , does this mean that the urdf file has no effect, right?(I'd like to confirm about it).

last , very glad to see this article research, Merry Christmas~

from dynamic_calibration.

Let me try to answer your questions

1. I haven't contacted UR for drive gain parameters, but I would write directly to technical support (ask an agent in China to refer you to technical support).
2. It might be too light. I would plot torques (currents) with and without load, the more they differ the better.
3. To avoid negative gains, set method of estimate_drive_gains(baseQR, method) function to 'PC-OLS'. Indeed the error is too big, to debug I would first assess the quality of filtering and play with filter parameters if necessary (filterData function). Sampling frequency shouldn't be an issue, but perhaps downsampling can help to filter and improve the quality of data.
4. Yes, URDF parser shouldn't be an issue

I am happy to know that you find our paper useful.

from dynamic_calibration.

Hello~, I've got it, now I have a further understanding, later I will re-do some actual experiments after finish the modification, I hope everything goes well, your answer is greatly appreciated, thank you~

from dynamic_calibration.