👋 Hi, I'm Yan Li.

This research focuses on humanoid robot locomotion using machine learning methods, particularly in the areas of reinforcement learning and imitation learning. The goal is to develop robust locomotion policies in simulation and then transfer them to a custom-built, 3D-printed humanoid robot. This pipeline bridges the gap between simulated learning and real-world application, enabling scalable and adaptable robot control.

We leverage vision-language models to automate labor-intensive tasks traditionally performed by humans. Our objective is to deploy computational solutions for forest surveillance, nursery monitoring, and disease control applications.

This project served as a bridge between my algorithmic work and real-world robot locomotion control. By integrating with the ROS protocol, it enabled reliable human tracking and medication recognition, allowing the robot to perform assistive tasks in dynamic environments.

This project aims to automate layout measurement in large-scale construction sites using a portable lofting device. Replacing traditional manual methods, the system combines laser projection with visual-depth sensing to accurately execute layout tasks. By interpreting building schematics in real time, it significantly reduces human error and improves efficiency—cutting layout time by up to 80%.

Empower non-robotics experts to train robots for tasks within their own domains through intuitive interfaces and adaptable learning frameworks.

This project was developed on top of the open-source autonomous driving framework Autoware. Given the limitations in computing resources and manpower, our research primarily focused on vehicle detection, lane detection, and instance segmentation. Additional efforts included LiDAR data processing, sensor calibration, as well as the development and modification of on-vehicle computing hardware and devices.

This project aims to assist visually impaired individuals in walking independently without the need for additional human support. Our approach leverages the convenience of wearable devices and their onboard sensors to collect data, which is then used to develop algorithms that enable effective navigation for the blind.