Reproduce “SPADE: Towards Scalable Path Planning Architecture on Actionable Multi-Domain 3D Scene Graphs” 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
Vignesh Kottayam Viswanathan, Akash Patel, Mario Alberto Valdes Saucedo, Sumeet Satpute, Christoforos Kanellakis, George Nikolakopoulos
Addressing the challenge of scalable robotic navigation in dynamic, uncertain environments, SPADE proposes a path planning architecture that utilizes actionable multi-domain 3D scene graphs. This framework achieves roughly 40% faster path computation through domain-aware graph subsampling and robustly adapts to changing traversability via a multi-level replanning strategy.
Key insights
- Prioritizing local layer planning and local geometric navigation, coupled with global route awareness, significantly enhances planning scalability in large-scale dynamic environments.
- Informed domain-aware subsampling effectively prunes multi-domain scene graphs, exposing only task-relevant edges to the planner, thereby reducing search space and improving efficiency.
- An iterative, multi-level replanning approach, which exhausts local solutions before escalating to global re-planning, enables robust adaptation to scene uncertainty with lower computational cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.