Lucy Lin

Hello! I'm a first year MS Robotics student at the Robotics Institute at Carnegie Mellon University, fortunate to be advised by Katerina Fragkiadaki. I am partially supported by the NSF Graduate Research Fellowhip (2025–2030).

Previously, I completed my B.S. at Oregon State University, where I worked with Weng-Keen Wong, Amir Nayyeri, and Fuxin Li.

Email  /  CV  /  Scholar  /  Twitter  /  Github

Research

I'm broadly interested in computer vision and its intersection with robot learning. Lately, I've been thinking about how to develop models that can ground their reasoning in spatial representations derived from perceptual input. I'm curious to see how this extends to embodied settings—specifically, how can we get agents to build persistent scene representations, reference their visual memory, and reason natively in space rather than text alone.

Outside of research, I also have a few hobbies.

A geometry-aware LMM that compresses multi-view observations into world-space tokens, attends with 3D Rotary PE, and grounds language via a query-based mask decoder for 3D grounding, segmentation, and VQA.

^*Equal contribution

We study biharmonic and k-harmonic distances on graphs, connect them to global topology and connectivity, and use them for edge centrality and graph clustering.

We extract RF device fingerprints from CNN+LSTM hidden-state patterns to improve open-set detection of unauthorized wireless transmitters.

Website template adapted from Jon Barron.