Cognitive Crowdsense: A Traffic-Psychology-Informed Framework for Pedestrian–Autonomous Vehicle Negotiation in Densely Populated Urban Environments

Pedestrian–vehicle interaction in dense urban traffic is governed not only by kinematics, but also by human risk perception, comfort (proxemics), and local interaction norms. This paper presents Cognitive Crowdsense, a traffic-psychology-informed decision augmentation for autonomous vehicles (AVs) operating in crowded environments. The framework is organized as five interlocking layers—Perceptual Empathy, Situational Recall, Emotional–Cognitive Response, Cultural Interpretation, and Ethical Judgment—and is implemented through three transparent engines: (i) a Mamdani neuro-fuzzy risk model that maps ambiguous social cues (e.g., separation gap, approach speed, and gaze/attention) to an interpretable social-risk score r*, (ii) an ethics-weighted right-of-way game that makes the safety–efficiency trade-off explicit via a single policy parameter ω, and (iii) calibrated Bayesian updates for sequential intent evidence, producing a crossing-intent probability π with reliability auditing. We follow a design-based validation blueprint by testing four pre-registered functional hypotheses (monotonic risk, equilibrium safety shift with ω, improved calibration, and reduced conflict outcomes) using a synthetic 10k scenario dataset informed by public benchmark statistics and closed-loop simulation rollouts. Beyond safety proxies, we report human-factors-relevant outcomes, including proxemic-violation counts, decision legibility (early-yield vs late hard-brake), and efficiency costs (delay and false-yield rate), to avoid the trivial “safe by stopping” failure mode. The results support the hypothesis set and provide an auditable pathway for aligning AV behavior with pedestrian expectations under cultural and ethical constraints.