Robotics · Embodied Intelligence · Vision-Language-Action Models
MSc Robotics & Artificial Intelligence graduate from University College London, working on generalizable robot learning, long-horizon embodied reasoning, and deployable VLA systems.
Research Profile
Research focuses on memory-augmented Vision-Language-Action systems that connect short-term perception with long-term task reasoning. Current work centers on temporally aware decision-making, multi-task robot learning, policy generalization, and robust deployment in complex real-world environments.
Research Experience
Research and engineering in embodied intelligence, VLA deployment, and general-purpose robot learning. Collaborated with USTC on the Robotic Chemist project, leading the design of a VLA-based system for long-horizon chemical experimentation with multi-scale memory and autonomous planning.
Conducted reinforcement learning research for embodied robotic control in BEHAVIOR-1K, designing scalable PPO training pipelines with reward shaping, multi-modal observations, and curriculum strategies for long-horizon household manipulation tasks.
Developed a deployable VLA framework combining in-context learning and parameter-efficient fine-tuning on a Qwen3-8B backbone. Proposed a multi-task LoRA design with shared low-rank subspaces, orthogonal task-specific heads, and cross-task regularization.
Selected as an Outstanding Project under the National College Student Innovation and Entrepreneurship Program. Contributed SLAM algorithms for robust indoor navigation and natural human-robot interaction in catering service robotics.
Education
MSc Robotics & Artificial Intelligence
BEng Robotics, Joint Program
BEng Computer Science and Technology
Selected Work
Ongoing work with USTC on long-horizon embodied manipulation using VLA models, multi-scale memory, and hierarchical decision-making.
Team member at Huawei London Research Centre, focusing on multi-task reinforcement learning and policy optimization within BEHAVIOR-1K.
Technical Skills
Python, PyTorch, Transformers, HuggingFace PEFT, Accelerate, ROS, MATLAB, Java, Git, Linux
PyBullet, RLBench, PyRep, CoppeliaSim, Gazebo
Computer vision, object detection, PEFT, MoE, in-context learning, reinforcement learning, VLA models
Chinese native, English fluent, IELTS 7.0
Honours
Beyond Research
Beyond the lab, a quieter rhythm takes shape: city walks, open coastlines, architecture marked by time, football as ritual, and music threading through papers, code, and experiments.
