This article presents a neuroscience-inspired computational model for virtual agents capable of autonomous learning through interaction with their environment. The proposed architecture is grounded in fundamental neurobiological principles such as sensory observation, selective attention, reward-based decision-making, and motivational persistence. The model is implemented and evaluated in a simulated environment where the agent must perceive, process, and act upon stimuli to achieve its goals. The results demonstrate that incorporating biologically plausible mechanisms enables adaptive behaviour and improves learning efficiency in dynamic virtual contexts.