Show HN: PyTorch-World v0.1.0: Build & Train World Models

3 points

6 months ago

I’m Param Thakkar, and my research interests center around Generative Modeling and World Models. Over the past months, I’ve been developing a PyTorch-based library aimed at making it dramatically easier to learn about world models, experiment with them, and build new ones. My goal is to create the most accessible, flexible, and extensible toolkit for anyone working in this space—from beginners exploring model-based RL to researchers pushing new architectures.

After releasing v0.0.1, which introduced initial support for the PlaNet world model, I’m excited to announce v0.1.0. This update expands the library significantly, introducing:

Broader environment support, including DMControl, OpenAI Gym, the Arcade Learning Environments, and more.

Full implementations of DreamerV1 and DreamerV2, with DreamerV3 currently in development.

Ongoing work to add additional world models, reusable components, and advanced environments such as Isaac Gym, ManiSkill3, and DeepMind Lab.

This release brings the project another step closer to a unified ecosystem for world-model research—one that accelerates innovation, lowers barriers to entry, and empowers the community to explore the frontier of model-based learning.

If this project interests you, consider starring the repository and joining me in contributing to its growth!

1 comment