We exploit scene graph properties to encode the extracted interaction features in a compact structure and segment the demonstration into blocks, streamlining the generation of Behavior Trees for robot ...