Ruei-Chi (Ricky) Lai

Hello! I'm Ricky, a Visiting Researcher at Texas A&M University, under the supervision of Prof. Zhengzhong Tu on robotics simulation and spatial understanding for MLLMs. Previously, I was a Research Assistant at the Vision Science Lab, National Tsing Hua University, where I was fortunate to be advised by Prof. Min Sun and mentored by Dr. Yi-Hsuan Tsai. I received my B.S. in Electrical Engineering & Computer Science from National Tsing Hua University in June 2025.

Currently, my research focuses on spatial understanding and reasoning for embodied AI, as well as efficient MLLMs.

Looking ahead, I aim to pursue a PhD starting in 2027 in robotics and MLLMs.

Email  /  CV  /  Google Scholar  /  Github  /  LinkedIn

Publications

Proposed online Geometry-aware token pruning method leveraging 3D information to identify overlapped redundant visual tokens in 3D scenes, achieving token reduction and +5.1 improvement on OpenEQA gain without fine-tuning.

A training-free framework for embodied QA that leverages question context to identify relevant objects and views in a scene, achieving consistent improvements across multiple 3DQA benchmarks without model fine-tuning.

Discovered that token pruning causes catastrophic performance drops in visual grounding due to misaligned position IDs. Our simple solution, GAP, recovers 90% of original grounding performance without additional training or computational overhead.

The website template is borrowed from Jon Barron, who created this amazing template.