I’m working with a quadruped robot to learn locomotion behavior with DRL using Ros2 and Drake in c++, and I want to apply domain randomization in my learning.
I would like some advice from the community to How I can change the physical properties of the robot (like mass, inertia, joint friction) and plane properties dynamically during the training.
Thank you.
One approach is that you can set the parameter in the context for each simulation. Here is a quick code for doing that
Notice calling
SetMassfunction, which takes the context as an input. You can callSetSpatialInertiaInBodyFrameandSetDampingmethods, both takecontextas an input.