Scaling Agent Learning via Experience Synthesis

arXiv 2025 38.2 method, system

TLDR

DreamGym synthesizes diverse experiences for scalable RL training using a reasoning-based experience model, replay buffer, and adaptive task generation.

Reasoning

The paper introduces a novel framework for experience synthesis, addressing scalability and cost in RL. Strengths include a unified approach and strong empirical results in both synthetic and sim-to-real settings. However, the abstract lacks explicit details on world model components, and the connection to the listed keywords is indirect, limiting direct relevance.

Read-first score

Read-first score 38.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 17.

Recency 8%

86.7

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

Methodology quality 25%

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

Topical relevance 42%

32.4

Matches configured research keywords against title, abstract, tags, and analysis text. matched=7

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: 267.

Keyword Scores

model-based reinforcement learning world model

world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Resources

Paper