A ROS 2-Based Robotic Platform for Mobile Occupant Sensing and Edge Perception in Buildings

Occupant-centric building operation requires timely information on occupant states and local indoor conditions, but fixed sensors provide limited spatial coverage, and wearables depend on user participation. This study develops a Robot Operating System 2 (ROS 2)-based wheeled mobile sensing platform for buildings. The main contribution is the integration of autonomous mapping and navigation, target approach, multisensor occupant-data acquisition, lightweight edge-based human detection, and return-to-dock operation on a Raspberry Pi 5. During operation, the robot patrols indoor locations, detects and approaches occupants, collects human and environmental data, uploads the data to a server, and returns to the charging dock. To support concurrent perception and navigation, YOLOv5n was compressed using structured channel pruning and multi-scale feature distillation. The compressed model reduced parameters and computation by approximately 53% and 61%, increased inference throughput from 7.5 to 16.5 frames per second, and enabled the robot to complete all 60 controlled trials across five locations and three postures. This study does not quantify HVAC energy savings or carbon-emission reductions. Instead, it validates a mobile sensing and edge-perception layer for future server-side thermal comfort inference, demand-responsive HVAC control, and evaluation of building energy and carbon performance.