A smart container for real-time load occupancy estimation using embedded neural inference

The increasing demand for sustainable urban delivery solutions has driven the adoption of cargo bikes due to their environmental benefits and adaptability to congested urban environments. These operations benefit from monitoring systems capable of estimating load occupancy (volume) to support logistical decision-making. This study presents a smart-container approach for real-time occupancy estimation using two Time-of-Flight (TOF) sensors and a compact neural model deployed on an ESP32-S3 microcontroller. TOF sensors generate distance matrices of the internal cargo space, which are processed to estimate occupied volume via the normalized FreeSpace target. In two inference stress tests, the system achieves R 2 = 0.929 and 0.923, with mean absolute error (MAE) = 0.044 on the normalized FreeSpace scale (0–1), equivalent to 8.1 dm 3 (4.4% of container capacity). The results support the feasibility of low-cost embedded inference for operational capacity checks in cargo-bike logistics.