Reproduce “Reasoning with Scene Graphs for Robot Planning under Partial Observability” 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
Saeid Amiri, Kishan Chandan, Shiqi Zhang
The Scene Analysis for Robot Planning (SARP) algorithm integrates dynamic global scene graph construction with probabilistic planning (POMDPs) to leverage semantic object relationships for biasing a robot's belief state. This approach reduced action costs by up to 50% and maintained high success rates in target search tasks within partially observable, object-rich environments, demonstrating improved scalability compared to traditional POMDPs.
Key insights
- Dynamically constructing a global scene graph from active robot perception provides a rich, domain-wide contextual understanding that surpasses single-image analysis.
- Leveraging learned semantic object relationships from scene graphs can effectively bias probabilistic robot planners, improving decision-making under partial observability without requiring hand-engineered knowledge.
- Applying context-aware belief biasing only upon target detection prevents error propagation and maintains policy quality, ensuring robust performance in environments with many objects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.