Event

ScheNav: Sketch-Guided Embodied Navigation Framework Released

1 reports · 1 sources

Guichen Huang and other authors released ScheNav, a closed-loop embodied navigation framework that translates coarse sketch guidance into exploration and stopping decisions. Its core is SchemaField, a learning module that fuses guidance with accumulated observations and predicts a spatial value field. The authors state that the framework learns to use sketch guidance entirely from simulated floor plans and transfers to unseen embodied environments. A supporting data pipeline converts 2D floor plans into navigation training environments, producing the Furnished-RPLAN dataset, which contains over 33k furnished layouts along with paired navigation episodes and noise-controllable sketches. The authors report that experiments improved navigation success rates and path efficiency under schematic diagrams, goal ellipses, and pointing directions, and that it tolerates the tested sketch perturbations and missing data.