Building intelligence into real systems
I work across multimodal AI, perception, and robotics systems. I enjoy the full path from a difficult model or systems problem to a system that feels coherent and dependable in practice.
Most recently, my research has explored interactive 3D scene understanding and robotics, including Act to See and a winning submission to the CVPR 2026 OpenSUN3D Challenge. Earlier, I built production LiDAR–camera tooling and pedestrian-tracking workflows at NIO.
I tend to work at three layers:
- Applied AI systems: model integration, reliability, evaluation, and deployment
- Efficient AI: language and vision models under latency, memory, and compute constraints
- Perception and robotics: sensor tooling, temporal models, 3D understanding, and action-conditioned representations
Fun fact: I have a twin brother, Chengkun (Charlie) Li, who is also actively engaged in artificial intelligence research.
Experience
Perception Engineer
- Participated in the development of a human-in-the-loop auto labeling system for 4D (3D + temporal) LiDAR-Camera data.
- Main contributor to a visualization and debugging tool for LiDAR and camera perception systems, adopted by engineers across multiple teams and featured in the company’s promotional video.
- Developed a web-based annotation tool, enabling human annotators to collect hundreds of hours of high-quality ground truth data for pedestrian movements in dense urban areas.
- Trained and deployed a sequence-to-sequence time series model for pedestrian tracking using point cloud, utilizing the collected high-quality data; this model now serves as an API for the annotation tool to provide real-time tracking results.
Graduate Studies, Electrical and Computer Engineering
- GPA: 4.0/4.0.
- Coursework: Neural Networks and Deep Learning, Introduction to Cloud Computing, Perception for Robotics, Cloud-Based Data Analytics
Graduate Studies, Electrical Engineering
- Coursework: Intelligent Systems, Generative Modeling, Computer Vision, Bayesian Machine Learning.