Appendix · World models

Back to core

Appendix: Index · Next: Memory

Overview#copy

World-model work aims to learn predictive, compressed representations of an environment (or domain) that support simulation, counterfactuals, control, and planning.

1) Self-supervised latent dynamics#copy

Key papers#copy

• World Models (Ha & Schmidhuber, 2018)
• Dreamer (Hafner et al., 2019)
• DreamerV2 (Hafner et al., 2020)
• DreamerV3 (Hafner et al., 2023)

Code repositories / projects#copy

• danijar/dreamerv3
• danijar/dreamer

2) Model-based control and planning#copy

Key papers#copy

• MuZero (Schrittwieser et al., 2019)
• Imagination-Augmented Agents (Racanière et al., 2017)

Code repositories / projects#copy

• google-research/google-research (MuZero reference implementations)
• werner-duvaud/muzero-general

3) Generative modeling as implicit simulators#copy

Key papers#copy

• Video Diffusion Models (Ho et al., 2022)
• Imagen Video (Ho et al., 2022)

Code repositories / projects#copy

• openai/guided-diffusion
• lucidrains/video-diffusion-pytorch