Insight from Training Robots

by

I am quickly responding to this video, with more notes later:

This is so fascinating, it explains how we learn and train our bodies to dance. It’s basically the same process I described here, which is what I learned at JMBT:

Training Your Body to Do Anything (within its physical capabilities)

Very exciting work.  Watching the video I believe the key missing component for people not experienced in dance is proprioception.  In order to train effectively we need: (a) direct conscious -brain access to and awareness of our body parts; and (b) direct conscious-brain access to the “error signal” that gets created and processed in our cerebellum as we move.  Therefore, work on Awareness Through Movement is critical in dance training: it builds the “missing” portions of the neural system that enable us to learn dance.

Reviewing it, interesting points are:

  • The interesting. difference between training human vs. animal (or this robot) bodies is… at least as adults, we get to decide on our reinforcement criteria and direct the process. Whereas animal training relies on an (external) human making that decision. Same with training children in dance. As children move into adulthood, they gain responsibility for the process, at least in dance training environments.
  • Proprioception was an important step in development of this robot: where the LLM was attached directly to the sensors and motor outputs. That is the key pice of neurological development that is lacking in most beginner students, and takes time to develop. Therefore, focusing on Awareness Through Movement (aka Feldenkreis) is *key* to dance training.
  • We get to choose our reinforcement goals; but one “natural” goal that usually comes in play, at least in dance movement, is *efficiency*: achieving the goal with less expended energy, less total movement, etc. I would like to see how the robot develops with efficiency as a component of the error signal.
  • I’m on the (autism) spectrum, and many of us have “clunky / clumsy” movement. That went away after 2 years of intense professional training, and I concluded that was because of intense development in my cerebellum. These robot experiments support that hypothesis.