The Trinity System for AGI Development: Lightweight Virtual Scene Base, Native Spatiotemporal Multimodal Perception, and Personalized Cognitive Evolution

The research and industrial deployment of Artificial General Intelligence (AGI) are currently facing three core bottlenecks. First, existing models center on data fitting and lack a closed-loop cognitive growth path based on temporal sequential scene interaction, imposing a clear ceiling on intelligence advancement. Second, multimodal perception technologies remain stuck in pixel-level statistical fitting, without establishing a native framework for spatiotemporal semantic cognition, which prevents deep understanding and natural interaction with real or virtual scenes. Third, the growth and alignment of AGI lack a controllable and customizable scenario-based environment. The traditional external strong-control paradigm is incompatible with the laws of intelligence evolution, making it difficult to balance the unleashing of technological potential and safety risk prevention. To address these pain points, this paper proposes a complete academic framework of the trinity system: Lightweight Virtual World Base – Native Spatiotemporal Multimodal Perception Interface – AGI Personalized Cognitive Evolution Core. It uses a low-energy, customizable lightweight virtual world as the native scene base for AGI growth and human-machine interaction, solving the closed-loop environmental problem of intelligence evolution. It adopts a multimodal perception system centered on a native spatiotemporal coordinate system as the cognitive and interaction interface, opening a two-way transmission channel between scene information and the intelligence core. It takes personalized perception grounded in rationality and accumulated temporal sequential scene experience as the path for AGI advancement, supported by a conflict adjudication mechanism between personalization and alignment targets, and a virtual-real world generalization adaptation mechanism, achieving a breakthrough from instrumental fitting to high-order cognition. This framework establishes a human-led, controllable, traceable, and two-way growth path for AGI, providing complete theoretical support and implementation solutions for the safe, sustainable, and large-scale deployment of AGI.