Reproduce “Scene Graph-Guided Proactive Replanning for Failure-Resilient Embodied Agent” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Che Rin Yu, Daewon Chae, Dabin Seo, Sangwon Lee, Hyeongwoo Im, Jinkyu Kim

A framework enhances autonomous robot resilience by proactively detecting potential failures via visual scene graph analysis before subtask execution. This approach, which compares observed environment states against successful demonstration graphs, enables more efficient replanning, achieving a 70% task success rate in simulation.

Key insights

  • Proactive failure detection, based on structured scene graph comparison, significantly outperforms reactive recovery methods in both task success rate and execution efficiency.
  • Integrating task-informed scene graphs provides critical relational and spatial context that general visual-language models often miss, leading to more robust failure anticipation.
  • Grounding Large Language Model reasoning in explicit scene graph differences enables the generation of highly accurate and human-preferred explanations for potential failures.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.