A Control-Centric Benchmark for Video Prediction

arXiv 2023 41.1 benchmark

TLDR

Proposes a control-centric benchmark (VP^2) for action-conditioned video prediction to evaluate models for robotic manipulation planning.

Reasoning

Strengths: addresses a gap in evaluating video prediction for downstream tasks, provides a comprehensive benchmark with multiple tasks and planning. Weaknesses: only simulated environments, no real-world validation; limited to robotic manipulation domain.

Read-first score

Read-first score 41.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 23.

Recency 8%

65.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2023

Methodology quality 25%

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

Reproducibility 25%

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

Topical relevance 42%

32.9

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: 154.

Keyword Scores

video world model

world model

world dynamics prediction

world simulator

generative world model

interactive world model

model-based reinforcement learning world model

Resources

Paper