Towards life-long adaptive agents: using metareasoning for combining knowledge-based planning with situated learning