Not exactly. Incorporating a world or domain model usually means taking pre-existing declarative knowledge in some form (e.g. from a semantic net) and using it

Not exactly. Incorporating a world or domain model usually means taking pre-existing declarative knowledge in some form (e.g. from a semantic net) and using it to aid learning. To quote the article, "Model-based reinforcement learning aims to address this issue by first learning a model of the environment’s dynamics, and then planning with respect to the learned model." So they've sped up learning from examples by learning a model first, but it's still learning from examples.