World Models Unlock Optimal Foraging Strategies in Reinforcement Learning Agents

arXiv 25.12 2025 41.4 method, application

TLDR

Model-based RL agents with learned world models converge to optimal foraging strategies aligned with the Marginal Value Theorem.

Reasoning

The paper presents a clear connection between world models and optimal foraging, with strong theoretical grounding in MVT. However, it lacks real-world validation and does not address scalability or broader applicability beyond simulated foraging tasks.

Read-first score

Read-first score 41.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 28.

Recency 8%

86.7

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

Topical relevance 42%

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

Methodology quality 25%

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

Reproducibility 25%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 339.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Demonstrates that artificial foragers with learned world models naturally converge to strategies aligned with the Marginal Value Theorem (MVT)
Shows that anticipatory capabilities, rather than reward maximization alone, drive efficient patch-leaving behavior
Reveals that model-based agents exhibit decision patterns similar to biological foragers, providing a foundation for explainable and biologically grounded AI

Methodology

The study employs a model-based reinforcement learning agent that learns a parsimonious predictive representation of its environment. The agent is evaluated in a patch foraging task and compared against standard model-free RL agents to assess alignment with the Marginal Value Theorem.

Key Results

Model-based agents converge to MVT-aligned optimal foraging strategies and exhibit decision patterns that mirror those of biological foragers, whereas model-free agents do not.

Resources

Paper