Robustness of Smart Parking Assignment Under Realistic Sensor Noise

Abstract

Searching for on-street parking remains a persistent urban challenge. Building on our prior coordinated parking framework (Cord-Approx), we upgrade its predictive engine with an attention-based multi-horizon model to better capture dynamic availability. We then stress-test the system under realistic imperfect sensing by introducing reduced sensor coverage (ρ) and false vacancy rate (φ), and show that, in the evaluated setting, coordinated assignment retains a clear advantage under noisy sensing. Even under severe noise, with 40% of free-spot detections missing (ρ=0.6) and a 5% false vacancy rate (φ=0.05), our system delivers a parking success ratio 2.3× that of competitors (non-users) and 51.9% lower search time.