This #rma work seems a lot similar to an idea I had for #robocup #humanoid walking, where you essentially have a base algorithm and them some #dynamic #adaptation algorithm: https://ashish-kmr.github.io/rma-legged-.. I still don't believe this is even close to the full potential of the method, and I think there are also some tricks being missed here... About to give the paper a quick read: https://ashish-kmr.github.io/rma-legged-..