Causal Object-Centric Models for Planning with Monte Carlo Tree Search

arXiv 2026 60.9 method

TLDR

COMET combines object-centric representations with Monte Carlo Tree Search for efficient model-based reinforcement learning.

Reasoning

The paper introduces a novel architecture that integrates a frozen object-centric encoder with a transformer world model and action-slot fusion, achieving strong early training performance across diverse benchmarks. However, the reliance on a frozen encoder and limited evaluation to early training stages may constrain generalizability.

Read-first score

Read-first score 60.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 50.

Recency 6%

100

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

Citation impact 18%

94.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.94705369

Topical relevance 29%

71.4

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 18%

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 471.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

interactive world model

generative world model

world simulator

video world model

Deep Analysis

Innovations

COMET algorithm combining Monte Carlo Tree Search with slot-structured latent space
Frozen unsupervised object-centric encoder paired with transformer-based world model
Action-slot fusion mechanism for binding actions to objects in slot transition prediction
Object-causal attention with learned per-slot relevance scores for policy and value heads

Methodology

COMET uses a frozen unsupervised object-centric encoder to produce slot representations, then a transformer-based world model with action-slot fusion to predict slot transitions. Policy and value heads employ object-causal attention modulated by learned per-slot relevance scores. The model is trained in a model-based RL setting with Monte Carlo Tree Search in the latent space.

Key Results

Across eight visually and dynamically diverse tasks from the Object-Centric Visual RL benchmark, ManiSkill, Robosuite, and VizDoom, COMET achieves a higher mean normalized score during the early stages of training compared to object-centric and monolithic baselines.

Resources

Paper