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

PIVOT-R: Primitive-Driven Waypoint-Aware World Model for Robotic Manipulation

NeurIPS 24 2024 48.3 method, application

TLDR

PIVOT-R is a primitive-driven waypoint-aware world model for robotic manipulation, achieving 19.45% improvement and 28x efficiency gain on SeaWave benchmark.

Reasoning

The paper introduces a novel waypoint-aware world model with an asynchronous executor, showing strong benchmark results. However, it lacks real-world robot validation and may oversimplify dynamics by focusing only on waypoints.

Read-first score

Read-first score 48.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 21.

Methodology quality 25%
80

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

Recency 8%
75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Reproducibility 25%
38

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

Topical relevance 42%
30

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

Field roles

Methodology anchor

Rank sensitivity

Stability: volatile; rank range: 347.

Keyword Scores

world model
10
world dynamics prediction
7
generative world model
2
world simulator
1
model-based reinforcement learning world model
1
interactive world model
0
video world model
0

Deep Analysis

Innovations

  • Primitive-driven waypoint-aware world model focusing on task-relevant waypoints
  • Waypoint-aware World Model (WAWM) with primitive action parsing and primitive-driven waypoint prediction
  • Asynchronous hierarchical executor (AHE) using different execution frequencies for different modules to reduce computational redundancy

Methodology

PIVOT-R consists of a Waypoint-aware World Model (WAWM) and a lightweight action prediction module. WAWM performs primitive action parsing and primitive-driven waypoint prediction, while the action prediction module decodes low-level actions. An asynchronous hierarchical executor (AHE) uses different execution frequencies for different modules to improve efficiency. The model is evaluated on the SeaWave benchmark against state-of-the-art open-source models.

Key Results

PIVOT-R achieves an average relative improvement of 19.45% over state-of-the-art open-source models across four levels of instruction tasks on the SeaWave benchmark. With AHE, execution efficiency increases 28-fold with only a 2.9% performance drop compared to synchronous execution.

Limitations

  • The asynchronous hierarchical executor introduces a 2.9% performance drop compared to synchronous execution
  • Evaluation is conducted only on the SeaWave benchmark, limiting generalization claims
  • Comparison is limited to open-source models, not including proprietary or closed-source systems

Resources

Paper

Tags