Task‐Oriented Hierarchical Environment Modeling via Spatial Belonging Relationships for Domestic Service Robots

ABSTRACT Efficiently performing household chores in dynamic environments remains a significant challenge for domestic service robots. While environment modeling is critical for task execution, existing approaches prioritize accuracy and efficiency over task relevance, limiting autonomy and long‐term robustness. This paper proposes a Hierarchical Environment Modeling for Domestic Service Robots (HEMDSR), emphasizing task‐related objects and their spatial belonging relationships (SBLs). Our framework includes: (1) a probabilistic SBL‐based model for adaptive object reasoning, (2) deep semantic representations and optimal viewpoint interest points for efficient object localization, and (3) a task‐oriented local 3‐D representation method to reduce memory overhead. Besides, inspired by human housework patterns, a systematic paradigm for HEMDSR‐based implementation is provided. Comparative experiments in simulation and real‐world environments demonstrate the validity and efficiency of the proposed solution. The results show that HEMDSR achieves path‐length performance competitive with human teleoperation while using substantially less memory during navigation.