BRICKS-WM: Building Reusability via Interface Composition Kinetics for Structured World Models

2026-06-15 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

BRICKS-WM (Building Reusability via Interface Composition Kinetics for Structured World Models) is a novel framework designed to enhance reusability in Model-based Reinforcement Learning (MBRL). It addresses the limitation of prevailing MBRL approaches that use monolithic latent dynamics, which tightly couple environment dynamics and necessitate retraining the entire world model when an agent changes. BRICKS-WM posits that global dynamics can be modeled as a composition of independent dynamical modules interacting through learned latent interfaces. As a minimal example, it factorizes the latent state into an actuated Agent module and an external Background module, connected by a learned interface. This approach enforces a functional separation in transition dynamics, ensuring the background dynamics remain independent of the agent's dynamics. Empirically, BRICKS-WM achieves control performance comparable to strong monolithic baselines when trained from scratch, and critically, enables the reuse of frozen background dynamics across different agents.

Key takeaway

For Machine Learning Engineers developing Model-based Reinforcement Learning systems, BRICKS-WM offers a crucial architectural shift. If your projects involve iterating on agents or transferring them across similar environments, adopting a modular world model like BRICKS-WM can drastically reduce retraining time and computational expense. You can reuse frozen background dynamics, accelerating development cycles and making MBRL more practical for complex, evolving applications. Evaluate its functional separation approach to improve system reusability.

Key insights

BRICKS-WM enables modular, reusable world models in MBRL by functionally separating agent and environment dynamics.

Principles

• Global dynamics can be composed of distinct interacting modules.
• Functional separation in transition dynamics enhances reusability.
• Background dynamics should be agnostic to agent dynamics.

Method

BRICKS-WM factors the latent state space into an actuated Agent module and an external Background module, bridged by a learned latent interface, enforcing functional separation.

In practice

• Reuse frozen background dynamics across different agents.
• Avoid retraining entire world models for agent changes.

Topics

• Model-based Reinforcement Learning
• World Models
• Modular AI
• Latent Dynamics
• Agent-Environment Interaction
• Reusability

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Learning and Reusing Policy Decompositions for Hierarchical Generalized Planning with LLM Agents — Dynamic policy learning with reusable components significantly boosts LLM agent performance and efficiency across diverse tasks.
• LeCun Shares: A Unified Toolkit for JEPA and World Model Research Is Finally Here — "stable-worldmodel" unifies disparate tools and data for world model research, streamlining development and reproducibility.
• ProPlay: Procedural World Models for Self-Evolving LLM Agents — ProPlay uses a procedural world model and graph to enable LLM agents to self-evolve by rehearsing and refining future task paths.
• Ambient Diffusion Policy: Imitation Learning from Suboptimal Data in Robotics — Ambient Diffusion Policy extracts useful features from suboptimal robot data by restricting its influence to specific diffusion times.
• The Sequence Knowledge #804: The Dreamer Trilogy: Inside Some of the Most Influential Papers in AI World Models — World Models address Model-Free RL's limitations in 3D environments by enabling agents to predict future states.
• Behavior-Invariant Task Representation Learning with Transformer-based World Models for Offline Meta-Reinforcement Learning — A novel OMRL framework uses behavior-invariant task representations and conservative value penalties to mitigate distribution shifts and enhance generalization in sparse-reward settings.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.