What is good PD parameters for Bolt?

@majid_khadiv

Hi Majid
I currently watch the expirment video you did on Bolt for thousand times.
The video is :a set of new tests on Bolt

It is really impressive that all joints of Bolt seems stiff enough to support dynamic walking,and not shocking.

I am currently trying to do the sim2real job on Bolt,but the joints are not stiff enough and may shock if increase the PD parameters.

Would you mind sharing the PD parameters using in the expirement,this can help me a lot.

BTW,do you do any filtering for joint velocity?Like low-pass filter etc…

Thanks very much!

Hi,

You can find simulation PD gains here. On real robot we used a fraction of P =[40.0, 40.0, 97.0], D = [0.26, 0.23, 0.16]. It is important to know, these parameters are tuned for our robot and reactive_planners code. I don’t think that these gains work for any algorithm.

Best,
Elham

1 Like

Thank you very much. I will give it a shot later.
BTW,when i tune the PD parameters,joints may shock because of bad parameters,this quiet often cause encoder error.Have you met this problem?

It is important to note the difference between the PD gains in mpc and rl, they are not the same thing. For mpc, the pd gains try to track a desired trajectory while in rl the D term is only damping (no desired velocity) and there is no feedforward term. To understand this better check the subsection “low level joint control” on page #9 of this paper https://arxiv.org/pdf/2406.01152. The gains Elham mentioned are for mpc, for rl with pd action on ODRI robots we always used kp=2 and kd=.1.

Best,
Majid