Create model with uncertainty (unkown paramters that can be updated online)

[Hongyu-ZHOU]

Hi,

I am trying to create a model of a quadruped robot from urdf file, and use that model for an adaptive control algorithm. In that case, is it possible to add uncertainty in the model generated? Essentially, I want to have a nominal model generated from the urdf file, augmented with some unknown parts (a vector of known basis function times a vector of unknown parameters.).

Thank you very much.

[ManifoldFR]

Hi, I moved this issue to the discussion tab if you don't mind!

I think what you can do is load up your model from the URDF file with some nominal parameters. This will yield a pinocchio.Model object in Python (pinocchio::Model in C++).

Then, you can write your own logic to play with the parameters you want to change in the Model object - gear ratio, friction, damping, inertias, joint placements, gravity. You can do that by accessing them on the object itself - we don't have pre-made functions for this, it's all left to the user.
You can keep using the same Data object when changing these properties. Make sure your simulator and/or control loop are pointing to these properties in the Model object for the physics.

One of the things I'm not sure you can change easily is the size of the vectors holding parameters, along with the structure of kinematic tree itself (adding/removing joints) and the operational frames. Then, you'd need to work with multiple Model instances and switch between them.

Good luck, have fun!