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

TransDreamerV3: Implanting Transformer In DreamerV3

arXiv 25.6 2025 56.3 method

TLDR

TransDreamerV3 enhances DreamerV3 with a transformer encoder, improving memory and decision-making in simulated RL tasks.

Reasoning

The paper clearly presents a novel architecture (transformer in DreamerV3) and provides experimental results on several tasks, showing improvements. However, it acknowledges limitations like issues in Minecraft and limited training, and lacks real-world validation.

Read-first score

Read-first score 56.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 33.

Recency 8%
86.7

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

Methodology quality 25%
80

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

Topical relevance 42%
47.1

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%
38

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 325.

Keyword Scores

world model
9
model-based reinforcement learning world model
9
world dynamics prediction
5
generative world model
4
world simulator
3
interactive world model
2
video world model
1

Deep Analysis

Innovations

  • Integration of a transformer encoder into the DreamerV3 architecture to enhance memory and decision-making
  • Application of transformer-based world models to reinforcement learning in complex environments

Methodology

TransDreamerV3 modifies DreamerV3 by implanting a transformer encoder to improve long-term memory and decision-making. The model is trained and evaluated on Atari-Boxing, Atari-Freeway, Atari-Pong, and Crafter tasks, with DreamerV3 as the baseline. Performance is measured by task-specific rewards or success metrics.

Key Results

TransDreamerV3 outperforms DreamerV3 on Atari-Freeway and Crafter tasks, but shows issues on the Minecraft task and suffers from limited training across all tasks.

Limitations

  • Performance issues observed in the Minecraft task
  • Limited training across all tasks, potentially affecting generalizability

Resources

Paper

Tags