Neural Scene Graphs (NSGs) offer a novel approach to modelling complex 3-dimensional (3D) environments by representing objects, attributes, and their spatial relationships within structured graphs. Combining graph-based scene representations with neural networks, NSGs provide robust and scalable solutions for scene synthesis, reconstruction, and manipulation tasks. This paper comprehensively reviews NSGs, exploring their architecture, methodologies, and applications across domains like Virtual Reality (VR), autonomous driving, robotics, and higher education. In educational contexts, NSGs enable immersive, interactive learning experiences by facilitating real-time rendering of dynamic 3D environments, making them ideal for simulations in fields such as medicine, architecture, and engineering.

The review delves into critical aspects of NSG development, including data acquisition, graph construction, and neural architecture optimisation, emphasising the potential for deployment in platforms like the Meta Quest 3 headset. Additionally, the paper discusses emerging techniques such as dynamic NSGs and AI-driven rendering optimisations, which address challenges in real-time performance and computational efficiency. By highlighting ongoing research and future opportunities, this paper underscores the transformative potential of NSGs in advancing 3D scene understanding, interactive applications, and educational tools that foster engaging hands-on learning.

Keywords: Neural Scene Graphs, 3D Scene Representation, 3D Scene Generation, Graph Neural Networks, Education.