This paper presents a distributed power control algorithm for multi-agent heterogeneous wireless networks to address queue stability and dynamic throughput maximization in Industrial Internet of Things (IIoT). The algorithm leverages the wireless channel's nature to approximate the centralized maximum-weight scheduling algorithm. It employs ubiquitous multi-protocol mobile devices as intermediaries to implement a concurrent dual-hop/multi-hop backhauling strategy, improving interoperability and facilitating data relay, translation, and forwarding from end IoT devices. The focus is on addressing large-scale network stability and queue management challenges. A long-term time-averaged optimization problem is formulated, incorporating end-to-end rate control, routing, link scheduling, and resource allocation to guarantee essential network-wide throughput. A real-time decomposition-based approximation algorithm is presented to ensure adaptive resource allocation, queue stability, and meeting Quality of Service (QoS) constraints with the highest energy efficiency. Numerical results show significant energy efficiency improvements across diverse traffic models, maintaining throughput requirements for both uniform and hotspot User Equipment (UE) distribution patterns. The work offers a comprehensive solution to enhance IIoT performance and address evolving challenges in industrial applications. Keywords: Maximum-weight scheduling, Queue stability, Distributed resource allocation, IIoT, Throughput utility maximization. The paper discusses the challenges of scaling existing IIoT using diverse IoT devices and proposes a solution leveraging ubiquitous mobile devices. It highlights the importance of dynamic cross-layer optimization design to address the intricacies of channel conditions, routing, transmission scheduling, device heterogeneity, and network stability in heterogeneous IIoT scenarios. The paper emphasizes the need for an efficient self-organizing multiaccess scheme to ensure seamless integration with next-generation mobile networks. It discusses the challenges of designing distributed Maximum-Weight Scheduling (MWS) algorithms and presents a learning-centric approach that does not assume a predefined wireless model. The algorithm uses bandit feedback, where each device receives feedback in the form of Acknowledgment or Non-Acknowledgment (ACK or NACK) after each packet transmission. The algorithm partitions the temporal horizon into epochs to achieve a distributed approximation.This paper presents a distributed power control algorithm for multi-agent heterogeneous wireless networks to address queue stability and dynamic throughput maximization in Industrial Internet of Things (IIoT). The algorithm leverages the wireless channel's nature to approximate the centralized maximum-weight scheduling algorithm. It employs ubiquitous multi-protocol mobile devices as intermediaries to implement a concurrent dual-hop/multi-hop backhauling strategy, improving interoperability and facilitating data relay, translation, and forwarding from end IoT devices. The focus is on addressing large-scale network stability and queue management challenges. A long-term time-averaged optimization problem is formulated, incorporating end-to-end rate control, routing, link scheduling, and resource allocation to guarantee essential network-wide throughput. A real-time decomposition-based approximation algorithm is presented to ensure adaptive resource allocation, queue stability, and meeting Quality of Service (QoS) constraints with the highest energy efficiency. Numerical results show significant energy efficiency improvements across diverse traffic models, maintaining throughput requirements for both uniform and hotspot User Equipment (UE) distribution patterns. The work offers a comprehensive solution to enhance IIoT performance and address evolving challenges in industrial applications. Keywords: Maximum-weight scheduling, Queue stability, Distributed resource allocation, IIoT, Throughput utility maximization. The paper discusses the challenges of scaling existing IIoT using diverse IoT devices and proposes a solution leveraging ubiquitous mobile devices. It highlights the importance of dynamic cross-layer optimization design to address the intricacies of channel conditions, routing, transmission scheduling, device heterogeneity, and network stability in heterogeneous IIoT scenarios. The paper emphasizes the need for an efficient self-organizing multiaccess scheme to ensure seamless integration with next-generation mobile networks. It discusses the challenges of designing distributed Maximum-Weight Scheduling (MWS) algorithms and presents a learning-centric approach that does not assume a predefined wireless model. The algorithm uses bandit feedback, where each device receives feedback in the form of Acknowledgment or Non-Acknowledgment (ACK or NACK) after each packet transmission. The algorithm partitions the temporal horizon into epochs to achieve a distributed approximation.