MAIA: Anticipating Interaction Anomalies in Contact-Rich Manipulation via Multimodal World Models

arXiv 2026 43.7 method

TLDR

MAIA uses a multimodal vision-action-force world model and the CRIA dataset to anticipate anomalies in contact-rich manipulation, improving task success via early warnings.

Reasoning

The paper introduces a novel anomaly anticipation framework with a real-world multimodal dataset and demonstrates clear gains in discrimination accuracy, lead time, and task success. Weaknesses include limited discussion of generalizability beyond the 18 evaluated tasks and reliance on abstract-level claims without visible implementation details.

Read-first score

Read-first score 43.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 48.

Methodology quality 15%

100

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

Recency 5%

100

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

Topical relevance 25%

68.6

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

Reproducibility 15%

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

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

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: 761.

Keyword Scores

world model

world dynamics prediction

interactive world model

world simulator

generative world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Multimodal vision-action-force world model that predicts visual and force responses of planned actions for anomaly anticipation
Lightweight anomaly head that predicts impending interaction anomalies from learned nominal interaction dynamics
CRIA dataset: a real-world multimodal dataset of 2,852 trajectories across 18 contact-rich manipulation tasks with frame-level anomaly annotations
Anomaly-aware action selection using anticipation predictions to improve task success without retraining the manipulation policy

Methodology

MAIA uses a vision-action-force world model to learn nominal interaction dynamics and predict visual and force responses of planned actions, paired with a lightweight anomaly head for detecting impending anomalies. Evaluation uses the introduced CRIA dataset, spanning 2,852 real-world trajectories across 18 contact-rich manipulation tasks, with experiments on both seen and novel tasks and comparison against representative baselines. Metrics include anomaly discrimination accuracy via bACC and median anticipation lead time, plus end-to-end task success when predictions are used for action selection.

Key Results

MAIA achieved bACC scores of 0.765 on seen tasks and 0.710 on novel tasks, with median anticipation lead times of 28.0 and 25.1 steps, respectively. Anomaly-aware action selection improved end-to-end task success from 52.6% to 61.5% without retraining the underlying manipulation policy.

Resources

Paper