Enabling self-orchestration on the Web of Things: A development framework

The exponential growth of the Internet of Things (IoT) and Web of Things (WoT) has revolutionized connectivity and data exchange. This growth poses challenges, particularly the complexity of managing many application components, and heightened resource consumption, crucial in constrained environments. To tackle these issues, a framework named Self-Orchestrated Web of Things (SOrWoT) is proposed for integrating Hierarchical Finite State Machine (HFSM)-based constructs into WoT, enabling application behavior modeling, breaking down complex systems, and further promoting changes in the architecting of components. This work proposes solutions for optimizing the reuse of Thing affordance architectings to demonstrate resource efficiency gains across scenarios. Namely, a mathematical optimization model and a Deep Q-Learning (DQL) agent are developed. The performance impact of network connectivity and components’ equivalence ratio is analyzed. Results show that the mathematical model consistently identifies optimal solutions by deterministically exploiting the feasible space, whereas DQL achieves near-optimal performance through adaptive learning, while offering greater flexibility and adaptability to dynamic environments. Across changing congestion conditions, DQL demonstrated resilience by exploring the environment and achieving higher rewards than in initial stages, highlighting its capacity for knowledge transfer in dynamic scenarios. Nonetheless, DQL proved more sensitive to action-space dimensionality and reward flattening, which influenced its convergence. Findings suggest that the combination of a high equivalence ratio with a low connectivity degree significantly boosts learning for better solutions. Our findings highlight the potential of the proposed framework in improving resource efficiency, reducing costs, and enhancing performance and scalability, by breaking down processes into smaller parts.