Set gains for right and left arms separately in the .ini
Issue #40
open
As we try to better tune our gains, we're seeing the right and left arms behave differently. Our right arm seems unable to handle integral gains as high as the left arm can do. Could the .ini files be modified so we can specify different gains for the right and left arms?
Comments (6)
-
-
reporter
Excellent, looking forward to the next release. While I'm asking for more gains to be exposed - could you please expose the joint teleop mode leg gains? We still get a lot of Atlas shaking when we lower him to the ground.
-
There is an issue with our stand-prep (that is before Atlas stands up): at that point we are not yet switched to our own PID controllers with integral leak and backlash compensation. During stand prep we are using the PID gains but send them to the BDI controllers. We have not gotten around to fix that. For now we probably could use a fraction of the gains during the stand up procedure that should get rid of the shaking (which we also see sometimes). I will look into that but this only affects the arms. Is there any other limb that you see shaking before Atlas stands up?
What do you mean by the joint teleop mode gains? We could potentially expose the gains used in position control mode but not the ones during the walking control mode.
-
reporter
We see Atlas shaking when he contacts the ground after being suspended in position control mode and when we switch from walking control to position control while standing on the ground. Position control mode is what I meant by joint teleop mode. We'd like to lower those leg gains.
We're not seeing any other shaking before Atlas stands up aside from the legs. When we lower him to the ground, there's a good chance he'll shake up and down once he makes contact. If he starts shaking, we have to raise him up a bit and try to lower him more gently.
-
I uploaded an ini file with the leg and back position control gains we currently use. We will test the code to load the file here shortly. Hopefully Doug can get a release out to enable that feature. Otherwise we can compile the code tomorrow when I get there so you are up to date.
-
- changed status to open
- Log in to comment
This will be possible with next release I updated ini file on develop branch. We also added some new stuff to our arm control:
Because of heavy velocity filtering our damping had so much delay that it did make things worse when we turned it on. The filtering is now much less so we were able to put some damping in the PIDs.
Since we did run into oscillations again because of the backlash in both shoulder joints we now actively reduce both P and D gain if we detect that we are moving in backlash. This results in the shaking dying down pretty quickly when it occurs. The gain reduction factor can be set in the ini file. If you wish to turn that function off just set the reduction factor to 1.0 and the gains will not change. We assume to be moving in backlash if the measured actuator force crosses zero.