External Human-Machine Interface based on Intent Recognition: Framework Design and Experimental Validation

Increasing autonomous vehicles (AVs) in transportation systems makes effective interactions between AVs and pedestrians indispensable. External human--machine interface (eHMI), which employs visual or auditory cues to explicitly convey vehicle behaviors can compensate for the loss of human-like interactions and enhance AV--pedestrian cooperation. To facilitate faster intent convergence between pedestrian and AVs, this study incorporates an adaptive interaction mechanism into eHMI based on pedestrian intent recognition, namely IR-eHMI. IR-eHMI dynamically detects and infers the behavioral intentions of both pedestrians and AVs through identifying their cooperation states. The proposed interaction framework is implemented and evaluated on a virtual reality (VR) experimental platform to demonstrate its effectiveness through statistical analysis. Experimental results show that IR-eHMI significantly improves crossing efficiency, reduces gaze distraction while maintaining interaction safety compared to traditional fixed-distance eHMI. This adaptive and explicit interaction mode introduces an innovative procedural paradigm for AV--pedestrian cooperation.