Performance Evaluation of an IoT-Enabled Mobile Robotic System for Smart Homes

Abstract

This paper introduces a cost-effective mobile robotic platform for automatic patrolling of multi-room apartments, which includes on-board vision-based occupancy detection using a model implemented on the Mobile Single Shot MultiBox Detector (MobileNet-SSD) on a Raspberry Pi 4B and load control using a relay module on ESP32. There is an independent kitchen gas-safety subsystem operating independently of the patrol cycle, characterized with 50 trials (mean response time 18.4 s, SD 4.3 s) and a false-alarm rate of 0.18 events/hour of cooking. Total hardware expense is 22,784 BDT (≈ USD 187, at 122 BDT/USD, 2025–2026 average). The detector attains an accuracy of 82% (95% Wilson score confidence interval (CI): 69%–90%) on a test set of 50 images, a precision of 95.7%, and a recall of 73.3%. For this two-room prototype, the latency for Google Firebase Realtime Database is lower than 500 milliseconds (ms) with a mean update time of 271 ms when no humans are detected and 338 ms when they are detected. A fan and lighting load is modeled to result in a net energy savings of 259 kilowatt-hours (kWh) per year, which translates into annual bill reductions of approximately 2,007 BDT under the Bangladeshi ascending-slab tariff. Simple payback is approximately 11.1 years for fan/lighting as built, reducing to 5.8 years with lower-cost compute hardware. The proposed system improves upon static passive infrared (PIR) sensors by detecting stationary occupants, ensuring privacy through on-device inference, integrating gas safety monitoring, and distributing the cost of a single sensing platform across multiple rooms. This work is a proof-of-concept prototype in support of the United Nations (UN) Sustainable Development Goals (SDGs) 7, 9, 11, 12, and 13.