Research on Large Model-Driven Multi-Source Fusion Algorithm for Visual Embodied Intelligence

Embodied intelligence aims to endow agents with the ability to perceive, make decisions, and interact with the physical world, while multi-source data fusion is a core challenge in enhancing their environmental understanding and task execution capabilities. This study proposes a large model-driven multisource fusion algorithm framework, which realizes an integrated closed-loop of perception-cognition-action through dynamic feature alignment and cross-modal reasoning. Experiments show that this framework significantly improves task success rates on benchmarks such as ALFRED and RoboTHOR, verifying the effectiveness of the method.