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

Planning in 8 Tokens: A Compact Discrete Tokenizer for Latent World Model

CVPR 26 2026 48.7 method

TLDR

CompACT tokenizer compresses observations into 8 tokens for faster, competitive planning in latent world models.

Reasoning

The paper presents a novel discrete tokenizer that significantly reduces computational cost for planning in world models, with strong empirical results showing orders-of-magnitude speedup. However, the abstract lacks explicit real-world evaluation and does not detail limitations or comparisons to other tokenization methods.

Read-first score

Read-first score 48.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 8%
100

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

Topical relevance 42%
50

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%
40

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

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 524.

Keyword Scores

world model
9
world simulator
6
model-based reinforcement learning world model
6
world dynamics prediction
5
generative world model
4
interactive world model
3
video world model
2

Deep Analysis

Innovations

  • CompACT discrete tokenizer compresses each observation into as few as 8 tokens, drastically reducing computational cost.
  • Enables orders-of-magnitude faster planning while maintaining competitive planning performance.
  • Provides a practical step toward real-world deployment of world models for decision-time planning.

Methodology

The paper proposes CompACT, a discrete tokenizer that compresses each observation into 8 tokens. An action-conditioned world model is built using this tokenizer to perform planning. The approach is evaluated on planning tasks, comparing against conventional tokenizers that encode observations into hundreds of tokens.

Key Results

CompACT achieves competitive planning performance with orders-of-magnitude faster planning compared to conventional tokenizers.

Resources

Paper

Tags