My long-term vision is to build a spatial intelligence ecosystem where cognitive intelligence, digital universe, and physical embodiments converge into a deeply synergistic, self-sustaining, and self-evolving cycle. This framework will empower machines not only to autonomously perceive and understand the physical world but also to proactively transform it. Driven by this synergy, AI will be deployed and scaled across increasingly complex spatial environments, ultra-long multimodal sensor streams, and diverse robotic morphologies, ultimately delivering transformative solutions to real-world challenges. Current research directions include:

1. Spatial Cognition: How can embodied agents perceive, represent, and reason about space and time like humans do?

2. Spatial Computing: How to create realistic digital twins from multimodal sensory streams without human supervision?

3. Spatial Robotics: How can we ground cognitive intelligence in real-world robots across different morphologies?

Recent representative works include: proposing H* Bench, the first in-the-wild benchmark for humanoid visual search; developing CLM, the first 3D Gaussian Splatting framework capable of billion-scale rendering on a single consumer-grade GPU; and designing Flex, an efficient multi-camera visual encoder for VLA-based autonomous driving.

During my Ph.D., I have published over 20 papers at top-tier conferences in computer vision, robotics, and machine learning (e.g., CVPR, ICCV, ECCV, NeurIPS, RSS, CoRL, ICRA), with several selected as Highlight, Spotlight, or Oral Presentation. My work has accumulated over 4,000 citations according to Google Scholar. I was awarded the NVIDIA Graduate Fellowship (10 recipients globally), the NYU Dean’s Fellowship, and was nominated for the NYU University-Wide Outstanding Dissertation Awards. Multiple research outcomes have been successfully transferred into technology at NVIDIA, with four U.S. patents filed or granted. For more information, please visit my homepage: https://yimingli-page.github.io/.