The Internet of Things (IoT) aims to create a global infrastructure that provides a variety of value-added services resulting from the interconnection of a large number of heterogeneous devices from different domains. Web of Things (WoT) enables to implement this vision by relying on the Web to virtualize and interconnect IoT objects in a transparent way. The IoT systems often operate in dynamic, unpredictable, and mobile environments with limited resources. Therefore, virtualized IoT objects on the Web need autonomous reasoning and decision-making capabilities to adapt their behaviors to their context. In this paper, we propose to extend virtual objects with reasoning mechanisms based on semantic models. The goal is to adapt their actions to the context of their environment. This extension is called Autonomous Avatar. It provides an artifact of a real-world entity on the Web while providing multiple features, such as standalone reasoning, context management, and collaboration capabilities. Based on the concept of autonomous avatar, we propose Fog based architecture for IoT Systems. This architecture is a pillar for defining IoT approaches for services delivery.