Yulun Zhang

Yulun Zhang

Ph.D. Student

Carnegie Mellon University

Biography

I am a first year Ph.D. student in Robotics Institute at the Carnegie Mellon University, advised by Jiaoyang Li. My research interests include human-robot collaboration, multi-robot coordination, evolutionary algorithms, and quality diversity optimization. As a long-term goal, my research focuses on bringing Quality Diversity Optimization and Evolutionary Optimization to Robotics, expanding their applicability and scalability.

Previously, I was a master/undergrad student majoring in Computer Science at the University of Southern California. I was working in Interactive and Collaborative Autonomous Robotics (ICAROS) lab at USC with Professor Stefanos Nikolaidis and Matt Fontaine. I was also working in Robotic Embedded Systems Laboratory (RESL) at USC with Ryan Julian and K.R. Zentner.

In addition, I was working in the USC Interaction Lab with Dr.Matt Rueben on Socially Assistive Robotics as well as Professor William Halfond’s group on record and replay tools for Android.

Here is my most recent CV (updated Oct 2022).

I am also enthusiastic about photography, especially scenery photography. Check out my portfolio and instagram for some of my works.

Interests
  • Quality Diversity Optimization
  • Evolutionary Computation
  • Human-Robot Interaction
  • Multi-Robot Coordination
  • Reinforcement Learning
Education
  • Ph.D. in Robotics

    Carnegie Mellon University

  • MSc in Computer Science, 2022

    University of Southern California

  • BSc in Computer Science, 2021

    University of Southern California

Projects

LDD Genie
A software tool for prosecution attorneys to perform logical document determination (LDD) electronically.
LDD Genie
Skill Builder: A Simple Approach to Continual Learning for Manipulation
We show that storing skill policies, careful pre-training, and fine-tuning with on-policy re-inforcement learning (RL), are sufficient for building a continual manipulation skill learner.
Skill Builder: A Simple Approach to Continual Learning for Manipulation
pyribs
A bare-bones quality diversity (QD) optimization library. QD algorithm is a powerful alternative compared to single-objective optimization because it is able to search for a diverse set of solutions according to user-defined metrics. In this library we provide implementations of recent QD algorithms such as MAP-Elites and Covariance Matrix Adaptation MAP-Elites (CMA-ME).
pyribs
Socially Aware, Expressive, and Personalized Mobile Remote Presence: Co-Robots as Gateways to Access to K-12 In-School Education
We propose a mobile remote presence robot system that enables children who cannot go to school in person for social and medical reasons to access in-school education.
Socially Aware, Expressive, and Personalized Mobile Remote Presence: Co-Robots as Gateways to Access to K-12 In-School Education
RERAN+
We found a time-sensitive issue of RERAN, a record and replay tool for Android published in 2013. In this project we propose a new scheduling mechanism for RERAN that resolve the time-sensitive issue.
RERAN+

Publications

Quickly discover relevant content by filtering publications.
(2022). Deep Surrogate Assisted MAP-Elites for Automated Hearthstone Deckbuilding. In GECCO ’22, July 9–13, 2022, Boston, MA, USA.

PDF Code Slides DOI

(2021). A Simple Approach to Continual Learning by Transferring Skill Parameters. Manuscript submitted for review.

PDF

(2021). Long-Term, in-the-Wild Study of Feedback about Speech Intelligibility for K-12 Students Attending Class via a Telepresence Robot. Proceedings of the 2021 International Conference on Multimodal Interaction (ICMI ’21), October 18–22, 2021, Montréal, QC, Canada.

PDF DOI

(2021). On the Importance of Environments in Human-Robot Coordination. In Robotics: Science and Systems 2021.

PDF Project

(2020). Towards Exploiting Geometry and Time for Fast Off-Distribution Adaptation in Multi-Task Robot Learning. In NeurIPS 2020 Workshop: Challenges of Real World Reinforcement Learning.

PDF Video