Awesome World Model Hub Papers · Datasets · Projects
← Back to papers

Learning To Explore With Predictive World Model Via Self-Supervised Learning

arXiv 25.2 2025 58.4 method

TLDR

Proposes an intrinsically motivated agent with a predictive world model via self-supervised learning, achieving superior performance on 18 Atari games.

Reasoning

The paper introduces a novel approach combining cognitive elements and self-supervised learning for world model-based exploration, with strong empirical results on Atari games. However, it lacks real-world validation and generalizability beyond simulated environments.

Read-first score

Read-first score 58.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%
70

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,result

Topical relevance 42%
62.9

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

Reproducibility 25%
30

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 272.

Keyword Scores

world model
9
world dynamics prediction
8
model-based reinforcement learning world model
7
interactive world model
6
world simulator
5
video world model
5
generative world model
4

Deep Analysis

Innovations

  • Proposes using several neglected cognitive elements to build an internal world model for intrinsically motivated agents
  • Employs self-supervised learning to enable exploration without hand-designed reward functions
  • Demonstrates emergence of cognitive skills (reactive and deliberative behaviors) in Atari games

Methodology

The methodology involves constructing an internal world model from cognitive elements and training an agent via self-supervised learning to generate intrinsic rewards. The agent is evaluated on 18 Atari games, and its performance is compared against state-of-the-art methods in both dense and sparse reward settings.

Key Results

The agent achieves superior performance compared to state-of-the-art in many test cases across dense and sparse reward environments, and cognitive skills emerge in games requiring reactive and deliberative behaviors.

Limitations

  • Evaluation is limited to 18 Atari games, which may not generalize to more complex or diverse environments
  • The specific cognitive elements used are not detailed in the abstract, leaving the approach partially unspecified
  • The claim of superior performance is qualified as 'in many test cases', indicating inconsistent improvements across all scenarios

Tags