Internet of Things (IoT) is a heterogeneous, mixed, and uncertain ubiquitous network, which has significantly affected the concept of wireless networking. A large number of wireless devices have connected through the IoT and shared large amounts of data. So, efficient routing and forwarding data packets from the wireless devices toward gateways, which have connected to the Internet, is one of the most important issues in IoT. This paper has focused on routing and forwarding data packets in IoT. Firstly, a learning automata-based cognitive framework has been applied to integrate cognition into IoT; because current IoT lacks intelligence and cannot satisfy the increasing application performance requirements, and also adding cognition into IoT equips it with a brain and high level intelligence. Then, a new routing and forwarding protocol, which benefits from cross-layer optimization between routing and Media Access Control (MAC) layer protocols, has been proposed. In the proposed protocol a network of variable action-set learning automata establishes a route between source nodes and a corresponding gateway, by making a directed acyclic graph toward the gateway. Then, using a set of learning automata, MAC layer protocol parameters are configured to properly forward data packets hop by hop. The proposed protocol has been named VLA-CR (Variable Action-set Learning Automata-based Cognitive Routing Protocol). Based on Martingale theorem, the convergence of VLA-CR has been proved. Finally, extensive simulation experiments have been conducted to show the performance of the proposed protocol. Simulation results show the superiority of VLA-CR over several existing routing protocol in terms of end-to-end reliability, end-to-end delay, power consumption, and routing time.

物联网（IoT）是一个异构，混合且不确定的泛在网络，已极大地影响了无线网络的概念。大量无线设备已通过IoT连接并共享大量数据。因此，从无线设备到已连接到Internet的网关的有效路由和转发数据包是IoT中最重要的问题之一。本文着重于物联网中路由和转发数据包。首先，基于学习自动机的认知框架已被应用于将认知集成到物联网中。因为当前的物联网缺乏智能，无法满足不断增长的应用程序性能要求，并且在物联网中增加认知能力使其具备了大脑和高级智能。然后，一个新的路由和转发协议，已经提出了从路由和媒体访问控制（MAC）层协议之间的跨层优化中受益的方法。在提出的协议中，可变动作集学习自动机的网络通过制作朝向网关的有向无环图，在源节点和相应网关之间建立路由。然后，使用一组学习自动机，将MAC层协议参数配置为逐跳正确转发数据包。拟议的协议已命名为VLA-CR（使用一组学习自动机，将MAC层协议参数配置为逐跳正确转发数据包。拟议的协议已命名为VLA-CR（使用一组学习自动机，将MAC层协议参数配置为逐跳正确转发数据包。拟议的协议已命名为VLA-CR（V良莠不齐动作集大号赚取甲基于utomata- Ç ognitive ř郊游协议）。基于Mar定理，证明了VLA-CR的收敛性。最后，进行了广泛的仿真实验，以证明所提出协议的性能。仿真结果表明，在端到端可靠性，端到端延迟，功耗和路由时间方面，VLA-CR优于几种现有路由协议。