Hybrid Decision Making Via Deep Learning for Integrated Visual Object Detection, Pose Estimation, and Energy Control in IoT Connected Multi-Robot Systems Using 5G Network

In dynamic, real-world environments, effective decision-making in Multi-Robot Systems (MRS) requires both robust perception and scalable inter-agent coordination. This work presents a unified perception and control framework that leverages a Faster Region-Based Convolutional Neural Network (Faster-RCNN) integrated with a Perspective-n-Point (PnP) optimization layer collectively referred to as the Faster-RCNN-Pose pipeline to enable accurate, low latency object detection and 6-DoF pose estimation across heterogeneous robotic agents. Unlike traditional decoupled pipelines, our end-to-end architecture performs simultaneous inference of object class, bounding box, and spatial orientation, thereby improving system level responsiveness and decision fidelity. By embedding this pipeline within an MQTT-based 5G communication infrastructure, robots can rapidly share spatial awareness, allocate tasks, and coordinate actions in real time. The proposed framework supports energy-aware navigation, multi-agent collaboration, and dynamic task reallocation, even as mission complexity and fleet size increase. Evaluations in open-environment deployments confirm that the system maintains high localization accuracy, low pose error, and consistent inter-agent coordination under motion, occlusion, and environmental variability. While communication latency remains a relevant constraint, especially in brokered architectures, the system includes latency compensation mechanisms and predictive control logic to preserve decision consistency. Overall, this work contributes a modular, scalable, and communication-resilient perception-to-decision pipeline for next-generation IoT-connected MRS, advancing the frontier of distributed collaborative robotics in unstructured, high-variability environments.