Fei Ni

Papers

Total Citations

H-Index

About

Fei Ni is a rising researcher at the intersection of generative AI and embodied intelligence, whose work is pioneering the use of diffusion models for robotic planning and decision-making. Their key research areas include offline reinforcement learning, robot manipulation, and multimodal reasoning. Ni’s major contribution lies in reimagining diffusion models not merely as generative tools but as adaptive, self-evolving planners. Their seminal paper, "AdaptDiffuser" (2023, 8 citations), introduces a framework that enables diffusion models to iteratively improve their own planning quality, overcoming the limitations of static, low-diversity training data—a critical step toward more robust autonomous agents. Building on this, Ni’s 2024 work, "Generate Subgoal Images Before Act" (4 citations), unlocks chain-of-thought reasoning in diffusion models for complex robot manipulation. By generating intermediate subgoal images from multimodal prompts (text and visual cues), this approach dramatically reduces task execution errors in long-horizon scenarios. Though early in their career, Ni’s ability to fuse generative modeling with hierarchical reasoning marks them as a transformative voice in robotics, with clear potential to shape how machines learn to plan and act in the physical world.