Breaking complex tasks into simpler sub-tasks organized in levels, where agents learn high-level strategies and low-level actions separately.