stable-worldmodel

About

stable-worldmodel is an open-source platform by galilai-group designed for reproducible world model research and evaluation. It provides a unified interface across three key stages: data collection, training, and model-predictive control (MPC) evaluation. Shipping with high-performance data format registries (like LanceDB) and common baselines (e.g., LeWM, DINO-WM), it streamlines model development. It also integrates a vast suite of standardized environments with customizable factors of variation, enabling effortless zero-shot generalization testing for robotics and control tasks.

Features

• Unified workflow for data collection, training, and MPC evaluation
• High-performance data registries including LanceDB and LeRobot adapter
• Standardized environments with customizable factors of variation
• Ready-to-use baselines and solvers (LeWM, DINO-WM)

Supported Platforms