Yunze's research primarily focuses on developing vision-centric multimodal models and AI agents. His research can be summarized into three consecutive thrusts: visual perception, multimodal reasoning, and agentic planning. The visual perception part focuses on object-centric perception systems and tokenization strategies in dynamic scenes; the multimodal reasoning part aligns strong perception models with language models for reasoning tasks; and the agentic planning part transforms the multimodal foundation models into AI agents with chain-of-thought capabilities to solve complex embodied or long-horizon tasks.