SAMPO: Scale-wise Autoregression with Motion PrOmpt for generative world models

arXiv 25.9 2025 62.5 method

TLDR

SAMPO combines visual autoregressive and causal modeling with motion prompts for generative world models, improving video prediction and control.

Reasoning

The paper introduces a hybrid framework that addresses spatial and temporal coherence in world models, with strong technical contributions like asymmetric tokenization and motion prompts. However, the abstract lacks explicit details on real-world evaluation and limitations, and the claimed improvements are not quantified.

Read-first score

Read-first score 62.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 55.

Recency 8%

86.7

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

Topical relevance 42%

78.6

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=experiment,metric

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: 378.

Keyword Scores

world model

generative world model

video world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

interactive world model

Deep Analysis

Innovations

Scale-wise autoregression with motion prompt (SAMPO) hybrid framework combining visual autoregressive modeling for intra-frame generation with causal modeling for next-frame generation
Integration of temporal causal decoding with bidirectional spatial attention to preserve spatial locality and support parallel decoding within each scale
Asymmetric multi-scale tokenizer that preserves spatial details in observed frames and extracts compact dynamic representations for future frames
Trajectory-aware motion prompt module that injects spatiotemporal cues about object and robot trajectories to focus attention on dynamic regions

Methodology

SAMPO is a hybrid framework that combines visual autoregressive modeling for intra-frame generation with causal modeling for next-frame generation. It employs temporal causal decoding with bidirectional spatial attention to preserve spatial locality and enable parallel decoding within each scale. An asymmetric multi-scale tokenizer handles observed and future frames differently, and a trajectory-aware motion prompt module injects spatiotemporal cues about object and robot trajectories to improve temporal consistency and physical realism.

Key Results

SAMPO achieves competitive performance in action-conditioned video prediction and model-based control, with 4.4× faster inference. It also demonstrates zero-shot generalization to unseen tasks and benefits from larger model sizes.

Resources

Paper