AdaReP:Adaptive Re-Planning under Model Mismatch for Neural World-Model Predictive Control

arXiv 2026 50.1 method, theory

TLDR

AdaReP adaptively adjusts replanning tolerance in neural world-model MPC, reducing computation by over 80% on real-world robots without modifying the model.

Reasoning

The paper presents a practical, training-free method to reduce computational overhead in world-model-based MPC by dynamically adjusting replanning frequency based on prediction mismatch and local dynamics sensitivity. Strengths include real-world robotic validation and significant query reduction; weaknesses are limited discussion of failure cases or comparison to other adaptive methods.

Read-first score

Read-first score 50.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 36.

Recency 6%

100

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

Citation impact 18%

95.1

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.9510524

Topical relevance 29%

51.4

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=none

Reproducibility 18%

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

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: 459.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Perturbation-based dynamic-regret framework analyzing the trade-off between replanning overhead and plan reuse under model mismatch
AdaReP: a training-free wrapper that adapts replanning tolerance online using current deviation from cached rollout and local sensitivity estimate
No modification to the learned world model or planner required

Methodology

AdaReP is a training-free wrapper that adapts the replanning tolerance online by computing the current deviation from the cached rollout and a local sensitivity estimate of the dynamics. It is evaluated across three settings: image-space planning, latent-space control, and real-world robotic manipulation, using a perturbation-based dynamic-regret framework to motivate the adaptive strategy.

Key Results

AdaReP substantially reduces planner-side computation while maintaining comparable task performance, including over 80% fewer queries on a 50-trial physical robot study.

Resources

Paper