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

AdaPower: Specializing World Foundation Models for Predictive Manipulation

arXiv 25.12 2025 61.1 method, application

TLDR

AdaPower adapts general world foundation models into specialist world models for robotic manipulation, achieving 41% improvement on LIBERO benchmarks without policy retraining.

Reasoning

The paper introduces novel adaptation techniques (TS-TTT and MP) that effectively bridge the gap between generative realism and control precision, with strong empirical results on LIBERO benchmarks. However, the evaluation is limited to simulated benchmarks, and real-world robotic validation is absent.

Read-first score

Read-first score 61.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 57.

Recency 8%
86.7

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

Topical relevance 42%
81.4

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 25%
50

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

Reproducibility 25%
30

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 519.

Keyword Scores

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

Deep Analysis

Innovations

  • Temporal-Spatial Test-Time Training (TS-TTT) for inference-time adaptation
  • Memory Persistence (MP) for long-horizon consistency
  • AdaPower framework that transforms general-purpose WFMs into specialist world models

Methodology

AdaPower is a lightweight adaptation framework that specializes general-purpose World Foundation Models (WFMs) into specialist world models using two novel components: Temporal-Spatial Test-Time Training (TS-TTT) for inference-time adaptation and Memory Persistence (MP) for long-horizon consistency. It is integrated within a Model Predictive Control (MPC) framework to empower pre-trained Vision-Language-Action (VLA) policies without retraining.

Key Results

AdaPower achieves over 41% improvement in task success rates on LIBERO benchmarks without policy retraining, while preserving computational efficiency and generalist capabilities.

Resources

Paper

Tags