Gaze is a natural and rapid non-verbal interaction mode, particularly well-suited for human–robot interaction in busy space environments. However, the space human–robot interaction based on gaze is still in its infancy. Therefore, this paper conducts a preliminary exploration in this area. Using the AAR-2, a free-flying astronaut-assistant robot, as the platform, we establish a gaze tracking database, construct a gaze tracking model based on an attention mechanism, and develop a human–robot interface. When the astronaut gazes at a control button on the interface, the corresponding control instruction is transmitted to the STM32 controller within the AAR-2 via radio frequency communication. Subsequently, the AAR-2 is propelled by ducted fans to perform the corresponding action. At the same time, the AAR-2 feeds back its operational state to the astronaut, thereby enabling space human–robot interaction. In the system, we achieve an effective gaze tracking model with high accuracy and implement an efficient image preprocessing method with high real-time performance. The experimental results demonstrate that the system can meet the actual requirements for accuracy and real-time processing.
Dai et al. (2026) studied this question.