Notes from Boston Dyanamics talk at NeurIPS 2020

In their 2016 NIPS talk they showed that they used mostly MPC + behavior templates. Seems still to be the same and not much ML?