Reproduce “Planning with Learned Object Importance in Large Problem Instances using Graph Neural Networks” 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
Tom Silver, Rohan Chitnis, Aidan Curtis, Joshua Tenenbaum, Tomas Lozano-Perez, Leslie Pack Kaelbling
Researchers at MIT CSAIL developed Planning with Learned Object Importance (PLOI), a method using Graph Neural Networks to identify a small, relevant subset of objects in large planning problems. This approach enables off-the-shelf planners to achieve orders-of-magnitude faster solutions and higher success rates in object-rich environments, including classical, probabilistic, and robotic task and motion planning.
Key insights
- A GNN can effectively learn a lifted, generalizable representation of object importance from small training problems to guide planning in much larger, unseen instances.
- Focusing planning attention on a dynamically determined small subset of objects, rather than the entire set, can dramatically improve planning efficiency and success rates without modifying the underlying planner.
- Automatically generated labels for 'sufficient objects' using a greedy reduction process are effective for supervised training of the object importance scorer, reducing manual effort.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.