RoboCraft: Learning to see, simulate, and shape elasto-plastic objects in 3D with graph networks

arXiv 2022 28.2 method, system, application

TLDR

RoboCraft uses particle representations and graph-network dynamics models with MPC to manipulate elasto-plastic objects from RGBD observations, learning from 10 minutes of real-world data.

Reasoning

The paper presents a clear, well-motivated system with both simulated and real-world experiments, showing strong empirical results for deformable object manipulation. However, the abstract lacks details on quantitative performance, baselines, and failure cases, and the method is specialized to object dynamics rather than general world modeling.

Read-first score

Read-first score 28.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 10.

Recency 8%

56.5

Uses a gentle age decay so recent papers surface without erasing older foundations. 2022

Methodology quality 25%

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

Reproducibility 25%

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

Topical relevance 42%

14.3

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

Candidate

Rank sensitivity

Stability: volatile; rank range: 303.

Keyword Scores

world dynamics prediction

world model

interactive world model

world simulator

model-based reinforcement learning world model

generative world model

video world model

Resources

Paper