Visual attention is a complex cognitive process that depends on both the internal state of an agent and the external environment. The agent must actively select parts of the sensory space to focus on to satisfy its internal needs. In this thesis we developed a deep learning model that uses visual and proprioceptive sensory input, directing attention to objects in the environment based on internal needs, thereby achieving complex goals. The model's development relies on the free-energy principle, utilizing predictive coding to predict sensory input and active inference for motor control and action execution. A generative model was implemented using a variational auto-encoder to predict visual sensory data. Extensive training and testing were conducted in a simulated environment to evaluate the model's performance. The results demonstrated that the model consistently directed attention to objects that satisfied its current needs and goals and could dynamically shift focus as the needs changed. With specific precision tunings, a behaviour similar to object permanence emerged. Furthermore, the model's architecture allows for the integration of additional sensory modalities, such as auditory or tactile inputs, which can enhance its ability to prioritize attention based on a broader range of sensory information.