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

Prismatic World Model: Learning Compositional Dynamics for Planning in Hybrid Systems

arXiv 25.12 2025 54.9 method, application

TLDR

PRISM-WM decomposes hybrid dynamics into composable primitives using MoE, reducing rollout drift for planning in continuous control.

Reasoning

The paper addresses a key challenge in model-based planning for hybrid systems with a structured MoE architecture and latent orthogonalization. Strengths include a principled approach to mode decomposition and empirical gains on simulated benchmarks. Weaknesses are the lack of real-world experiments and reliance on simulated environments only.

Read-first score

Read-first score 54.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Methodology quality 18%
90

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

Recency 6%
86.7

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

Topical relevance 29%
64.3

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

Citation impact 18%
55.3

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.55279448

Reproducibility 18%
30

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

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 388.

Keyword Scores

world model
10
world dynamics prediction
9
model-based reinforcement learning world model
8
generative world model
7
world simulator
6
interactive world model
4
video world model
1

Deep Analysis

Innovations

  • Decomposition of complex hybrid dynamics into composable primitives using a structured architecture
  • Context-aware Mixture-of-Experts (MoE) framework with a gating mechanism for implicit physical mode identification
  • Latent orthogonalization objective to ensure expert diversity and prevent mode collapse

Methodology

PRISM-WM employs a context-aware Mixture-of-Experts (MoE) framework where a gating mechanism implicitly identifies the current physical mode, and specialized experts predict the associated transition dynamics. A latent orthogonalization objective is introduced to maintain expert diversity and prevent mode collapse. The model is trained on continuous control benchmarks and evaluated as a substrate for trajectory optimization algorithms like TD-MPC.

Key Results

Experiments on high-dimensional humanoids and multi-task settings demonstrate that PRISM-WM reduces rollout drift and provides a high-fidelity substrate for trajectory optimization, indicating its potential as a foundational model for model-based agents.

Limitations

  • Reliance on implicit mode identification may fail in highly ambiguous or unseen hybrid dynamics
  • Performance on real-world robotic systems has not yet been demonstrated, as evaluation is limited to simulation benchmarks

Resources

Paper

Tags