Mind Dreamer: Untethering Imagination via Active Causal Intervention on Latent Manifolds

arXiv 2026 51.2 method

TLDR

Mind Dreamer improves model-based RL by using active causal intervention to generate non-continuous latent jumps, achieving 1.67x speedup over DreamerV3.

Reasoning

The paper introduces a novel method to address historical tethering in model-based RL with theoretical contributions like quadratic discount and spectral gap analysis. However, empirical evaluation is limited to the DeepMind Control Suite, and real-world physical experiments are absent.

Read-first score

Read-first score 51.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 49.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 29%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.6999539

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=none

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierBridge

Rank sensitivity

Stability: volatile; rank range: 427.

Keyword Scores

world model

model-based reinforcement learning world model

generative world model

world dynamics prediction

interactive world model

world simulator

video world model

Deep Analysis

Innovations

Active Causal Intervention to transcend Markovian continuity in model-based RL
Reformulation of discovery as minimization of global Relay Expected Free Energy
Adversarial generator for initial states creating non-continuous latent jumps to epistemic blind spots
Relay Value Function and Relay Uncertainty Function for credit assignment across spatial ruptures
Proof that uncertainty propagation across discontinuities necessitates quadratic discount γ^2
Theoretical variance-minimizing importance sampler that expands manifold's spectral gap

Methodology

Mind Dreamer (MD) is a model-based reinforcement learning framework that uses an adversarial generator to sample initial states from a distribution p_gen, bypassing historical data. It derives Relay Value and Uncertainty functions to propagate pragmatic and epistemic value via Bellman-style backups, with a quadratic discount factor for uncertainty across discontinuities. The approach is evaluated on the DeepMind Control Suite against DreamerV3.

Key Results

MD achieves a 1.67× average speedup over DreamerV3 on DeepMind Control Suite, with an 8.8× speedup on sparse-reward tasks.

Resources

Paper