Multimodal TOR interfaces for L3 autonomous vehicles: a tradeoff analysis of driver trust, cognitive load, and takeover efficiency

The study focused on designing effective Takeover Request (TOR) interfaces to balance driver trust, cognitive load, and takeover efficiency in Level 3 automated driving. This study used a driving simulator to evaluate 16 multimodal interfaces that combined four visual designs, two auditory cues, and two haptic feedback states. Driving performance, subjective feelings, and eye-tracking metrics were measured, and a comprehensive ranking was established using the entropy weight method (EWM). The interface combining auditory and haptic cues with contextual visual information exhibited the best overall performance (EWM > 0.927). Auditory cues significantly improved driver trust (average improvement of 1.29 points, p < .001) and enhanced takeover efficiency by reducing reaction time by 0.13 s (p < .001), demonstrating a 1.5 times the trust-building effect of haptic cues. Haptic cues effectively reduced cognitive load (DALI:Δ = −3.79, p < .001) and compensated for processing delays in visually complex scenarios. For example, when combined with contextually enhanced visual information, it could reduce reaction time by 0.13 s (p < .001). Optimal TOR performance stems from strategic function allocation across modalities: utilizing auditory and haptic channels to issue emergency alerts while simultaneously using the visual channel for contextual information. This multimodal functional complementarity approach effectively balances trust, cognitive load, and efficiency, providing theoretical support for designing human-machine interfaces in next-generation autonomous vehicles.