What reward structure would you design to teach a robot to untangle overlapping items in a sorting bin for Berkshire Grey?
