A Survey of Object Goal Navigation: Datasets, Metrics and Methods

Object Goal Navigation (ObjectNav) aims at directing an agent to a specified target object within an unseen scene. This task integrates advanced techniques, including visual perception, semantic prior learning, and visual navigation, which makes it a challenging and emerging task within the Embodied AI field. In this survey paper, we offer a comprehensive review of widely-used datasets and corresponding evaluation metrics for ObjectNav. Additionally, we provide an in-depth review and analysis of recently developed methods. Finally, we present discussions on potential future research directions.