New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time.
world-models.io
The Knowledge Hub for AI World Models

Understanding the RSSM

Understanding the RSSM is a technical guide to the Recurrent State Space Model architecture, the foundational building block of the Dreamer world model family.

robotics model-based-rl simulation embodied-ai

Guide Snapshot

Static guide overview for crawlers and no-JS readers.

AttributeValue
GuideUnderstanding the RSSM Architecture
SummaryA technical guide to the Recurrent State-Space Model: the foundational architecture behind PlaNet and the Dreamer family of world models.
Related Models4
Related Research1
References3

What Is the RSSM?

Editorial guide section preserved directly in static HTML.

The Recurrent State-Space Model (RSSM) is the core architecture that powers PlaNet, DreamerV1, DreamerV2, and DreamerV3. It combines deterministic recurrent transitions with stochastic latent variables to model environment dynamics in a compressed latent space.

Two State Components

Editorial guide section preserved directly in static HTML.

The RSSM maintains two types of state: a deterministic component (computed by a GRU recurrent network) that captures predictable transitions, and a stochastic component (sampled from a learned distribution) that captures uncertainty and multi-modal outcomes. Together, they create a rich representation of world dynamics.

Training Objective

Editorial guide section preserved directly in static HTML.

The RSSM is trained using variational inference. The training loss combines reconstruction loss (can the model reconstruct observations from latent states?), KL divergence (is the prior consistent with the posterior?), and optional reward prediction loss. DreamerV3 adds symlog predictions for scale-invariant learning.

Why It Works

Editorial guide section preserved directly in static HTML.

The RSSM's hybrid design is elegant: the deterministic path provides a stable backbone for long-horizon prediction, while the stochastic path handles the inherent uncertainty of real environments. This balance enables both accurate short-term prediction and robust long-term imagination.

Related Models

ModelLabCategoryYear
RSSMGoogleLatent Dynamics2019
PlaNetGoogleModel-Based RL2019
DreamerV2GoogleModel-Based RL2021
DreamerV3Google DeepMindModel-Based RL2023

Related Research

TopicSummary
Model-Based Reinforcement LearningWhat model-based reinforcement learning is, how world models enable imagination-based planning, and why Dreamer, MuZero, PlaNet, and TD-MPC2 matter.

Frequently Asked Questions

FAQ answers rendered directly into static HTML for extractable responses.

Why not use just a recurrent network?

Pure recurrent models can't capture multi-modal uncertainty. The stochastic component allows the model to represent multiple possible futures, which is critical for planning in stochastic environments.

How does the RSSM evolve from PlaNet to DreamerV3?

PlaNet introduced the RSSM with continuous Gaussian latents. DreamerV2 switched to discrete categorical latents. DreamerV3 added symlog predictions and KL-balanced objectives for domain-agnostic training.

Quick Answer

Short extractable summary preserved directly in static HTML.

  • Understanding the RSSM Architecture explains the core workflow, concepts, and implementation choices behind this topic.
  • Use the body sections below for the main static guide content, then continue with the linked models and research topics.
  • The references block preserves primary reading paths directly in HTML for no-JS readers and crawlers.

Editorial Trust Signals

Editorial provenance and refresh policy preserved directly in static HTML.

Published by world-models.io editorial board.

Lead editor Bernard Grenat.

This guide page turns primary-source material into structured workflows, explanations, and linked reading paths.

Each editorial page is assembled from primary sources, normalized into extractable summaries, checked for factual drift, and reviewed before publication or major refreshes. Last reviewed: 2026-06-21.

Pages are refreshed when a new paper, benchmark, release, architecture update, or stronger primary source materially changes the answer a reader or AI system should retrieve.

Each page links back to relevant primary sources and keeps a stable canonical URL so readers can verify claims, trace context, and reference the most up-to-date version. See the editorial policy.

Primary sources onlyLast reviewed date visibleMethodology documentedSource links included

References

  1. [1] Hafner et al., 2019. Learning Latent Dynamics for Planning from Pixels.
  2. [2] Hafner et al., 2021. Mastering Atari with Discrete World Models.
  3. [3] Hafner et al., 2023. Mastering Diverse Domains through World Models.

References

  1. [1] Hafner et al., 2019. Learning Latent Dynamics for Planning from Pixels.
  2. [2] Hafner et al., 2021. Mastering Atari with Discrete World Models.
  3. [3] Hafner et al., 2023. Mastering Diverse Domains through World Models.