Toward cognition-aware augmented reality for construction safety: a systematic review and dual-loop framework

• Reviews 128 studies on AR and cognition in construction safety. • Proposes a dual-loop framework for cognition-aware AR safety systems. • Shows current AR systems are strong in site-state support but mostly open-loop. • Finds cognitive-state research is strongest in sensing and inference. • Outlines practical pathways to adaptive AR for safer construction work. Augmented reality (AR) has been increasingly applied in construction safety to provide site-specific information and improve hazard identification, communication, training, and task guidance. However, most existing AR systems remain open-loop and focus mainly on technological functions while paying limited attention to workers’ cognitive states. This may reduce the effectiveness of AR support in dynamic and cognitively demanding construction environments. Augmented cognition (AugCog), which combines real-time cognitive-state sensing with adaptive human–system interaction, offers a promising pathway for addressing this limitation. To clarify this potential, this paper proposes a dual-loop framework that integrates a site-state loop and a cognitive-state loop and uses it to guide a systematic review of two complementary literature streams. The review covers 128 peer-reviewed studies, including 28 AR-based construction safety studies and 100 cognition-focused construction safety studies. Results show that current AR research has made strong progress in site-state sensing, visualization, and task support, but remains largely open-loop with respect to worker cognition. In contrast, the cognition-focused literature shows substantial development in cognitive-state sensing and inference, with more limited but emerging progress in adaptation and evaluation. Direct integration of the two loops within a single AR safety system remains rare. This study contributes an evidence-informed framework for understanding how AR and AugCog can be integrated in construction safety and identifies practical pathways for developing adaptive, cognition-aware AR systems that better support workers’ situation awareness and reduce cognitive overload.