Back to Modules
advancedAdvanced

Applications and Extensions

Discover real-world applications and recent advances in World Models research.

25 min read
3 references

Applications and Extensions

Real-World Applications

World Models have been applied to various domains:

Robotics

  • Sim-to-Real Transfer
  • Manipulation Tasks
  • Tool use and dexterous manipulation

Autonomous Driving

  • Scene Understanding
  • Decision Making
  • Safe trajectory planning

Game AI

  • Procedural Content Generation
  • NPC Behavior
  • Adaptive opponents

Recent Extensions

DreamerV1/V2/V3

The Dreamer series extends World Models with actor-critic training and discrete latent spaces.

PlaNet

Planning Network combines latent dynamics model with model predictive control.

IRIS

Imagination with Reinforcement learning In Simulation uses transformer-based world models.

Genie

Generative Interactive Environments learns from unlabeled video data.

Future Directions

  • Foundation World Models
  • Hierarchical World Models
  • Multimodal World Models
References
Academic papers and resources

Dream to Control: Learning Behaviors by Latent Imagination

Danijar Hafner et al. (2020)

paper

Mastering Diverse Domains through World Models

Danijar Hafner et al. (2023)

paper

Genie: Generative Interactive Environments

Jake Bruce et al. (2024)

paper