GBPP Enables Fast Robot Base Pose Selection via Two-Stage Learning

GBPP uses a PointNet++ encoder and MLP to predict optimal robot base placement from single RGB-D snapshots. A two-stage curriculum combines low-cost distance-visibility auto-labeling with high-fidelity simulation refinement to outperform proximity and geometry-based baselines without requiring full motion optimization.