As the proportion of autonomous vehicles (AVs) gradually increases in future transportation, pedestrian-vehicle interactions face new challenges. Existing research primarily focuses on vehicle physical safety or explicit human-machine interfaces (eHMIs), while studies on how deceleration cues influence pedestrians’ psychological experiences remain further investigated. Our study, from a human-centered perspective, systematically explored the combined effects of different deceleration onset times (reflected by time to arrival of vehicles-TTA if not decelerate) and initial vehicle speeds on pedestrians’ perceived safety, comfort, and deceleration detection time. Utilizing a within-subjects design of 4 (speed: 30/40/50/60 km/h) × 7 (TTA: 1.0–4.0 s, at 0.5-s intervals), 45 participants interacted with AVs in virtual road scenarios. Results show that although deceleration with smaller TTA was easier to detect; pedestrians perceived safety and comfort increased with longer TTA. Based on the findings, we propose a “2.5-s critical threshold” principle: with a fixed stopping distance of 3.5 m before pedestrian, it is recommended that autonomous driving systems start decelerating at least 2.5 s in advance (TTA ≥ 2.5 s). The research findings offer a quantified design guideline for autonomous driving motion planning algorithms.