Reproduce “VBM-NET: Visual Base Pose Learning for Mobile Manipulation using Equivariant TransporterNet and GNNs” 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
Lakshadeep Naik, Adam Fischer, Daniel Duberg, Danica Kragic
VBM-NET enables mobile manipulators to select optimal base poses for object grasping directly from top-down visual observations, integrating navigation costs without explicit object state estimation. The method achieves a planning time of 0.14 seconds and an 81.3% success rate in simulation, demonstrating successful sim-to-real transfer.
Key insights
- Orthographic top-down projections provide a consistent, global, and scale-preserving scene layout, which is crucial for visual-based base pose planning and integrating navigation cost considerations.
- Exploiting SE(2) spatial equivariance with a TransporterNet significantly improves sample efficiency and robustness in visually identifying feasible base poses compared to non-equivariant models.
- Graph Neural Networks effectively represent and process a variable number of candidate base poses, enabling the selection of an optimal pose within a dynamic action space.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.