Samuel Li

I am a M.S. in Robotics student at Carnegie Mellon University advised by Prof. Katia Sycara. I recieved my B.S. in Mathematics & Computer Science from University of Illinois Urbana-Champaign, where I worked on computer vision with Prof. Yuxiong Wang and machine learning for climate prediction with Prof. Ryan Sriver.

I am currently researching zero-shot 3D vision (semantic decomposition and generative reconstruction) and robot manipulation guided by LLMs infused with symbolic knowledge. Outside of research and classes, I like to play tennis, cook, backpack, and spend time with my dog and cat.

Email  /  Resume  /  GitHub  /  LinkedIn

Research

ShapeGrasp: Zero-Shot Task-Oriented Grasping with Large Language Models through Geometric Decomposition

We develop a novel and efficient zero-shot, task-oriented grasping pipeline constructing a symbolic graph from monocular RGB+D input for fine-grained, shape-based LLM reasoning.

Geometric Shape Reasoning for Zero-Shot Task-Oriented Grasping

We propose a lightweight zero-shot, task-oriented grasping approach utilizing LLMs for part-level semantic reasoning over geometric decompositions.

Let Me Help You! Neuro-Symbolic Short-Context Action Anticipation

We develop a novel modification to the transformer architecture for short-context action anticipation, enabling human-robot collaboration in real-world experiments.

YouTubePD: A Multimodal Benchmark for Parkinson’s Disease Analysis

We introduce the first publicly available Parkinson’s disease analysis benchmark and demonstrate the generalizability of our developed models to real-world clinical settings.

Skillful Prediction of UK Seasonal Energy Consumption based on Surface Climate Information

We show how winter climate and energy demand values can be predicted two months in advance using surface climate information.

Projects

Enhancing Sample Efficiency via Affordance-Based Exploration

With Yuchen Zhang, Yunchao Yao, and Yihan Ruan. We solve an optimal control problem on a robotic arm to accurately throw an object to a goal location.

Enhancing Sample Efficiency via Affordance-Based Exploration

With Yunchao Yao, we leverage affordance understanding in foundation models for efficient, safe, and aligned task-conditioned exploration and learning for robotic manipulation.

Design and source code from Leonid Keselman's Jekyll fork of Jon Barron's website