TL;DR
A unified framework that standardizes the entire workflow of world model research—data collection, training, and evaluation—to enhance reproducibility.
Key features
Unified Interface: Handles data collection, model training, and evaluation with model-predictive control (MPC) through a single API.
Standardized Environments: Supports various simulation environments like PushT.
Baselines & Solvers: Provides reference implementations of common baselines and planning solvers (e.g., CEMSolver).
Flexible Data Formats: Automatically detects and loads various data formats such as Lance.
When to use it
Use this when researching world models or evaluating model performance in new environments, to save time on experimental setup and reproducibility, allowing focus on the model itself.