Quantifying the Robustness of Smart Street Parking Assignment to Sensor Noise

Abstract

Searching for street parking remains a major source of congestion in dense cities. Prior studies either assume perfect sensing of free spots or study sensing separately, leaving the robustness of smart parking assignment under sensing errors unclear. We address this using a Madrid-calibrated city-scale simulator and a replacement-based sensor-noise model parameterized by sensor coverage ρ and false vacancy rate ϕ, where missed detections hide real free spots and false vacancies report occupied spots as free. We extend our Cord-Approx strategy by embedding arrival-aware predictions into the Hungarian assignment cost and by rerouting participants after false vacancy encounters. Under perfect sensing, the parking success ratio rises to 82.2%, up from 77.54% in our prior work and closer to the 85.32% Cord-Oracle upper bound. Even at ρ=0.6, ϕ=0, participants still achieve 74.2% success and 8.92 minutes of search time, versus 31.1% and 19.45 minutes for competitors. This coordinated advantage remains visible in all five Madrid districts in a representative noisy setting and remains positive across the evaluated grid, with a practical operating boundary around ϕ ≈ 0.15.