Evaluating an AI-enhanced power wheelchair obstacle alert system: Effects of multimodal warning design on warning-response behavior

Accessible and inclusive transport requires mobility technologies that support safe, efficient, and understandable interaction for users with diverse sensory, cognitive, and physical needs. Power wheelchairs are a critical form of personal mobility, yet collisions with obstacles remain a common safety concern. Although intelligent wheelchair systems can detect hazards, evidence on effective warning communication remains limited in time-critical navigation scenarios. This study evaluated an AI-enhanced power wheelchair obstacle alert system in which a You Only Look Once (YOLO)-based pedestrian detection pipeline and depth-based trigger logic were used to activate multimodal warnings through auditory and vibrotactile cues. Twenty-four able-bodied university participants completed wheelchair navigation trials in which warning design varied by vibration configuration, auditory signal type, and information type. Vibration configuration included no vibration, seatback vibration, and wrist vibration; auditory signal type included non-speech and speech cues; and information type included instructional warnings that indicated the required avoidance action and informative warnings that indicated the obstacle location. Results showed that instructional warnings produced shorter reaction times and were rated as more useful and satisfying than informative warnings. Seatback vibrations were associated with shorter reaction times and higher usefulness ratings than audio-only warnings. Non-speech auditory cues produced shorter reaction times than speech cues, but this effect was significant only when no tactile cue was provided. Decision-making accuracy did not differ significantly across warning conditions. These findings suggest that direct action-oriented warnings, especially when paired with seatback vibrotactile feedback, may support faster responses to power wheelchair obstacle warnings. The study contributes to accessible transport design by identifying multimodal interface strategies that may improve human-machine communication in intelligent mobility devices.