Doorway-Aware Frontier Scheduling for Indoor Mobile Robots Using RGB-D Geometric Screening and Temporal Semantic Anchors
TL;DR
The results support the feasibility of using geometrically screened, temporally maintained doorway cues to modify frontier scheduling; direct traversable-area coverage, detector accuracy, runtime, and real-robot performance remain outside the present evidence.
Abstract
Indoor mobile robots must select navigation goals while operating with incomplete maps, but raw visual semantic detections can be too unstable to influence planning directly. This paper presents a doorway-aware frontier-scheduling framework that combines RGB-D geometric screening, temporal semantic-anchor maintenance, an incremental semantic-support layer, and anchor-dependent frontier-utility reweighting. Doorway candidates enter the planning loop only after depth-validity, local-plane, and geometric-consistency checks; accepted observations are then fused into persistent anchors whose confidence decays after missed updates. In ten paired same-condition simulations, the proposed configuration increased the mean semantic-support activation ratio from 5.11 × 10−4 to 1.447 × 10−3 and was higher in nine of the ten paired runs (two-sided exact sign test, p = 0.0215). The mean number of completed navigation goals increased from 0.10 to 2.50 per run, and all episodes terminated normally under the fixed evaluation protocol. Single-run demonstrations in two auxiliary layouts confirmed closed-loop executability but were not treated as statistical generalization evidence. The results support the feasibility of using geometrically screened, temporally maintained doorway cues to modify frontier scheduling; direct traversable-area coverage, detector accuracy, runtime, and real-robot performance remain outside the present evidence.