Using CALM for Sim-to-Real Transfer in Humanoid and Quadruped Robots

[Luckykantnayak]

Dear CALM authors and users,

I am interested in applying CALM for sim-to-real transfer of learned policies on humanoid and quadruped robots and would appreciate any insights from your experience.

In particular, I would like to understand:

Are there recommended practices or configurations when using CALM for sim-to-real transfer?

What are the main bottlenecks or failure modes you have encountered when transferring CALM-trained policies to real hardware?

[tesslerc]

sorry for the late response on this.
we haven't tried applying CALM in sim-to-real.

however there shouldn't be any limitation in terms of the method.
to the best of my knowledge, sim-to-real adaptation is orthogonal to the control problem (assuming the policy doesn't demand inputs the robot is incapable of providing).