Page 27 is particularly interesting, shows like how model-based RL would work with Dyna.

Dyna is model-based RL with a learned model. AlphaGo Zero is model-based RL with a known model.

It corrects itself and adapts to potential inaccuracies or changes in the world through a continuous interplay of learning from real experiences and dedicated strategies for exploration.

Citation

(Moerland et al., 2020)

Moerland, T. M., Broekens, J., Plaat, A., & Jonker, C. M. (2020). Model-based Reinforcement Learning: A Survey. https://arxiv.org/pdf/2006.16712