Reproduce “CuriousBot: Interactive Mobile Exploration via Actionable 3D Relational Object Graph” 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
Yixuan Wang, Leonor Fermoselle, Tarik Kelestemur, Jiuguang Wang, Yunzhu Li
CuriousBot, a mobile robotic system, leverages an actionable 3D relational object graph and LLM-based planning to perform interactive exploration by manipulating objects and uncovering occluded spaces in complex household environments. The system achieved an 82% success rate in diverse tasks, outperforming VLM baselines in object recovery and graph accuracy.
Key insights
- Explicit 3D relational object graphs are more effective for interactive mobile exploration, providing superior topological reasoning and action grounding compared to implicit reasoning from 2D visual observations by VLMs.
- Integrating Visual Foundational Models for perception and Large Language Models for planning with a structured, actionable 3D representation enables more informed decision-making for comprehensive autonomous exploration.
- The system demonstrates robust generalization across various object categories, object relations, and scene layouts, highlighting its adaptability to complex real-world environments beyond constrained settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.