University Seminar Paper
The algorithm DreamerV2 presents a recent breakthrough in the field of model-based reinforcement learning, reaching state-of-the-art perfor- mance on the popular ATARI Learning Environment benchmark. This seminar paper gives an overview over model-based reinforcement learn- ing, shows the predecessors of DreamerV2 and explains the algorithm and neural-network architecture in detail. The results, possible applications and the features of model-based reinforcement learning are discussed.
YouTube Video with an overview over DreamerV2 (part of the seminar)
Full seminar paper: PDF
The original DreamerV2 Paper is written by Danijar Hafner, Timothy P. Lillicrap, Mohammad Norouzi and Jimmy Ba.
Official Repository
Official Project Website
Errata: Please create an issue on this repository or contact me directly!
