PLUME: Probabilistic Latent Unified World Modeling and Parameter Estimation for Multi-Finger Manipulation

arXiv 2026 63.9 method, application

TLDR

PLUME jointly learns a belief over physical parameters and dynamics for dexterous manipulation, enabling zero-shot sim-to-real transfer.

Reasoning

The paper presents a novel world model that handles parameter uncertainty through online inference, with strong empirical results on simulated and real tasks. However, the evaluation is limited to a few specific manipulation tasks and only one real-world experiment.

Read-first score

Read-first score 63.9, 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%

93.9

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.93931012

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,evaluation

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 431.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

interactive world model

generative world model

world simulator

video world model

Deep Analysis

Innovations

Probabilistic Latent Unified world Modeling and parameter Estimation (PLUME) that jointly learns a belief over physical parameters and system dynamics conditioned on those parameters
Novel learning framework enabling online parameter inference for efficient alignment of the world model to true dynamics without re-training or fine-tuning
Latent space representation that jointly captures multiple qualitatively different physical parameters and rewards for informed planning

Methodology

PLUME learns a latent world model that evolves a belief over physical parameters (e.g., object shape, pose, friction) and models system dynamics conditioned on those parameters. The model is trained in simulation with known parameters and deployed with online inference to align with true dynamics, using a latent space that also encodes rewards for planning. Evaluation is performed on simulated screwdriver turning, valve turning, bucket lifting, and disk flicking tasks, plus a hardware screwdriver turning task, with comparisons to offline reinforcement learning and world-model-augmented behavior cloning baselines.

Key Results

The simulation-trained policy achieves successful zero-shot transfer to a hardware screwdriver turning task and outperforms state-of-the-art offline reinforcement learning and world-model-augmented behavior cloning baselines across all evaluated tasks.

Limitations

Method may require careful design of the latent space to effectively capture multiple physical parameters and their interactions
Online parameter inference might not converge quickly or accurately for all possible parameter variations or task scenarios
Evaluation is limited to a specific set of manipulation tasks (screwdriver turning, valve turning, bucket lifting, disk flicking) and one hardware demonstration, so generalization to other dexterous manipulation tasks is not yet demonstrated

Resources

Paper