# stable-worldmodel | GalilAI-group | Knowledge Base | Instagit

A platform for reproducible world model research and evaluation

GitHub Stars: 1.3k

Repository: https://github.com/galilai-group/stable-worldmodel

---

## Articles

### [Troubleshooting Solver Convergence and Planning Horizon Selection in Stable-WorldModel](/galilai-group/stable-worldmodel/troubleshooting-solver-convergence-and-planning-horizon-selection)

Struggling with Stable-WorldModel solver convergence Stop planning horizon mismatches Learn how to tune horizon and solver hyperparameters for reliable results.

- Tags: how-to-guide
- Published: 2026-05-30

### [Execution Flow of World.collect vs World.evaluate in stable-worldmodel](/galilai-group/stable-worldmodel/understanding-world-collect-vs-world-evaluate-execution-flow)

Understand the difference between World collect and World evaluate execution flow in stable-worldmodel. Learn how they record data and compute metrics for effective RL experimentation.

- Tags: internals
- Published: 2026-05-30

### [How to Implement Custom Data Normalization and Observation/Reward Encoding for World Models](/galilai-group/stable-worldmodel/implementing-custom-data-normalization-and-obs-rew-encoding)

Learn to implement custom data normalization and obs/rew encoding for world models in galilai-group/stable-worldmodel. Extend the Transformable protocol and register your scaler for seamless integration.

- Tags: how-to-guide
- Published: 2026-05-30

### [Setting Up Environment Variation Parameters for Robustness Evaluation of World Models](/galilai-group/stable-worldmodel/setting-up-environment-variation-parameters-for-robustness-evaluation)

Explore robust world models by setting up environment variation parameters. Customize visual and physical factors for domain randomization and OOD testing with Stable WorldModel.

- Tags: how-to-guide
- Published: 2026-05-30

### [Implementing Categorical CEM for Discrete Action Spaces in Stable-WorldModel](/galilai-group/stable-worldmodel/implementing-categorical-cem-for-discrete-action-spaces)

Learn how to implement categorical CEM for discrete action spaces in Stable-WorldModel. Optimize action sequences efficiently without gradients using Gumbel-max sampling and elite refitting.

- Tags: how-to-guide
- Published: 2026-05-30

### [Understanding World Model Rollout and Embedding Caching in Stable-WorldModel](/galilai-group/stable-worldmodel/understanding-world-model-rollout-and-embedding-caching)

Learn how galilai-group/stable-worldmodel optimizes rollout and caching with vision transformers for efficient model-predictive control. Discover embedding caching techniques.

- Tags: internals
- Published: 2026-05-30

### [How to Implement Action Repeat and Frame Stacking Wrappers for World Models](/galilai-group/stable-worldmodel/implementing-action-repeat-and-frame-stacking-wrappers)

Learn to implement action repeat and frame stacking wrappers for world models using the galilai-group/stable-worldmodel library. Streamline your environment modifications efficiently.

- Tags: how-to-guide
- Published: 2026-05-30

### [Supported Dataset Formats for Stable-WorldModel: HDF5, Lance, LeRobot, and Folder Layout](/galilai-group/stable-worldmodel/dataset-formats-hdf5-vs-lance-vs-lerobot-vs-folder-structure)

Explore supported dataset formats for Stable-WorldModel: HDF5, Lance, LeRobot, and folder layouts. Discover the flexible data handling capabilities of stable_worldmodel.

- Tags: api-reference
- Published: 2026-05-30

### [How to Implement Lagrangian and PGD Solver Callbacks for Constrained MPC in Stable‑WorldModel](/galilai-group/stable-worldmodel/implementing-lagrangian-and-pgd-solver-callbacks-for-constrained-mpc)

Learn to implement Lagrangian and PGD solver callbacks for constrained MPC in Stable-WorldModel. Extend solver constructors, invoke callbacks at each step, and aggregate histories for advanced control.

- Tags: how-to-guide
- Published: 2026-05-30

### [EnvPool Parallel Environment Execution and Masking in Stable-WorldModel: A Technical Deep Dive](/galilai-group/stable-worldmodel/understanding-envpool-parallel-environment-execution-and-masking)

Master EnvPool parallel environment execution and masking in Stable-WorldModel. Achieve high throughput for your world models by understanding selective masking and pre-allocated buffers.

- Tags: deep-dive
- Published: 2026-05-30

### [Implementing Custom Environment Wrappers with Pre‑wrappers and Extra‑wrappers in stable‑worldmodel](/galilai-group/stable-worldmodel/implementing-custom-environment-wrappers-with-pre_wrappers-extra_wrappers)

Learn to implement custom environment wrappers in stable-worldmodel using pre-wrappers for pre-processing and extra-wrappers for post-processing. Enhance your model's data pipeline effectively.

- Tags: how-to-guide
- Published: 2026-05-30

### [GPU Memory Optimization Strategies for Large World Models: 8 Techniques from Stable-WorldModel](/galilai-group/stable-worldmodel/gpu-memory-optimization-strategies-for-large-world-models)

Discover 8 GPU memory optimization strategies for large world models. Train huge models on single 24 GB GPUs using mixed-precision inference, Torch compile, and more from Stable WorldModel.

- Tags: performance
- Published: 2026-05-30

### [Implementing Online Learning with Iterative World Model Updates in Stable-WorldModel](/galilai-group/stable-worldmodel/implementing-online-learning-with-iterative-world-model-updates)

Learn to implement online learning with iterative world model updates in Stable-WorldModel. Enhance agent planning by alternating environment interaction and gradient-based TD-MPC2 model updates.

- Tags: how-to-guide
- Published: 2026-05-30

### [Understanding the MegaWrapper Preprocessing Pipeline and Image Transforms in stable-worldmodel](/galilai-group/stable-worldmodel/understanding-megawrapper-preprocessing-pipeline-and-image_transform)

Explore the MegaWrapper preprocessing pipeline and image transforms in stable-worldmodel. Standardize environments to a pixel-based RL interface with efficient image preprocessing.

- Tags: deep-dive
- Published: 2026-05-30

### [Using MergeDataset and ConcatDataset for Multi‑Source Data Composition in Stable‑WorldModel](/galilai-group/stable-worldmodel/using-mergedataset-and-concatdataset-for-multi-source-data-composition)

Master multi-source data composition in world models using MergeDataset and ConcatDataset. Join columns horizontally or concatenate episodes vertically for seamless integration with trainers.

- Tags: how-to-guide
- Published: 2026-05-30

### [How to Implement Custom Cost Functions for World Model Planning in Stable WorldModel](/galilai-group/stable-worldmodel/implementing-custom-cost-functions-for-world-model-planning)

Learn to implement custom cost functions for world model planning by subclassing Costable and overriding the criterion method in Stable WorldModel. Preserve gradients for optimized planning.

- Tags: how-to-guide
- Published: 2026-05-30

### [Configuring Action Spaces for Model-Based Planning Solvers in Stable‑WorldModel](/galilai-group/stable-worldmodel/configuring-action-spaces-for-model-based-planning-solvers)

Configure action spaces for model-based planning solvers in Stable-WorldModel using configure() and PlanConfig.action_block for optimal solver initialization and dimensionality.

- Tags: how-to-guide
- Published: 2026-05-30

### [LeWM vs PreJEPA vs GCRL: Which World Model Architecture Should You Choose?](/galilai-group/stable-worldmodel/lewm-vs-prejepa-vs-gcrl-choosing-world-model-architecture)

Compare LeWM, PreJEPA, and GCRL world models. Select LeWM for MPC, PreJEPA for multimodal JEPA, and GCRL for stochastic goal-conditioned policies. Find the best fit for your project.

- Tags: comparison
- Published: 2026-05-30

### [Implementing Custom Visual Wrappers for World Models: Occlusion, Chroma‑Key, and Moving‑Patch Guide](/galilai-group/stable-worldmodel/implementing-custom-visual-wrappers-occlusionwrapper-chromakeywrapper-movingpatchwrapper)

Learn to implement custom visual wrappers for world models using OcclusionWrapper, ChromaKeyWrapper, and MovingPatchWrapper. Easily add visual effects without changing environment logic.

- Tags: how-to-guide
- Published: 2026-05-30

### [CEM Solver Configuration Parameters for World Models: Understanding num_samples, horizon, and receding_horizon](/galilai-group/stable-worldmodel/understanding-cem-solver-configuration-parameters-num_samples-horizon-receding_horizon)

Master CEM solver configuration for world models. Understand num_samples, horizon, and receding_horizon to optimize action sampling and prediction for better planning.

- Tags: how-to-guide
- Published: 2026-05-30

### [How to Implement Custom Reward Shaping Wrappers for World Model Training in stable-worldmodel](/galilai-group/stable-worldmodel/how-to-implement-custom-reward-shaping-wrappers-for-world-model-training)

Learn to implement custom reward shaping wrappers for world model training in stable-worldmodel. Subclass gym Wrapper to modify rewards for improved agent learning.

- Tags: how-to-guide
- Published: 2026-05-30