Open traversaro opened 4 years ago
diegoferigo
commented
4 years ago I want to bump this issue since it is affecting all our simulations. Is there available any (even unofficial) list of recently identified friction parameters? I don't think that these parameters should be extremely accurate, having at least the right order of magnitude would help a lot. I fear that setting kv=1 is quite a high number.
cc @robotology/iit-dynamic-interaction-control
traversaro
commented
4 years ago Good point, I think that the one that most recently worked on this is @nunoguedelha . To clarify the historic setting, a quite high damping was actually choose to reduce the numerical noise when simulating the model in Gazebo with the ODE physics engine, but as we discussed in https://github.com/robotology/icub-models/issues/33 it is quite important that for the future we try to keep strictly separate the "nominal" model parameters, and any simulator-specific hack.
cc @pattacini @Nicogene
Since when the iCub URDF and Gazebo models have been created, they have been always arbitrary values for the viscous and coulomb joint friction parameters (see https://github.com/robotology/icub-model-generator/pull/86).
While the values of viscous and coulomb friction that best approximate a real joint depend on a lot of factors (such as the precise transmission and motor models, and how much the transmisssions are lubrificated, etc etc) and change from robot to robot and also on the calibration data, I think we should be able to at settle on parameters that have the correct order of magnitude, and use them by default on our iCub models.