CURRENT
USC
Ph.D. Student · LiraLab
08/25 → PRESENT
cs.phd / usc.liralab
Bo-Ruei Ray
Huang
黃柏睿 · PhD @ USC LiraLab · AI, robotics, science
I am a second-year Ph.D. student in Computer Science at the University of Southern California, advised by Erdem Bıyık in the LiraLab.
I received my Bachelor's degrees in Electrical Engineering and Computer Science from National Taiwan University, where I was advised by Shao-Hua Sun in the NTU Robot Learning Lab. In 2024 I visited MIT CSAIL, working with Jiayuan Mao, Xiaolin Fang, Leslie Pack Kaelbling, and Joshua B. Tenenbaum. Previously, in the summer of 2023, I worked with Yuk L. Yung at Caltech GPS.
// 02
trajectory
04 nodes
VISIT
MIT
Visiting Student · CSAIL
08/24 → 01/25
INTERN
Caltech
SURF · GPS
06/23 → 08/23
UNDERGRAD
NTU
B.Sc. CS + EE
09/20 → 01/25
// 03
news.log
tail -n 4
[2026.07]
Gave a hands-on tutorial on
RL with human feedback
at the
AIxHRI Summer School
in Nancy, France.
TUTORIAL
show 4 oldershow less
[2024.10]
KALM won
Best Paper at the
Workshop on Language and Robot Learning @ CoRL 2024.
★ BEST PAPER
[2024.10]
Our paper
Diffusion Imitation from Observation
was accepted by
NeurIPS 2024. See you in Vancouver.
NEURIPS · 24
[2024.08]
Fortunate to visit
MIT CSAIL,
hosted by
Jiayuan Mao in
Joshua B. Tenenbaum's and
Leslie Pack Kaelbling's group.
MIT · CSAIL
// 04
research_focus
My research aims to make robots that learn to help humans in their daily lives — especially in long-horizon, open-world tasks where instructions are ambiguous and the environment is unstructured.
I work across reinforcement learning, imitation learning, planning, and representation learning, with a recent focus on leveraging foundation models to bootstrap robust, generalizable behaviors.
// 05
publications
05 entries
Beyond the End-Effector: Learning Whole-Arm Contact Manipulation from Human Demonstrations
WACL learns whole-arm contact from bare-hand human videos and pairs it with language-generated path constraints, so a robot can push, press, and nudge with its arm while its hand stays busy with another task.
DEXOP: A Device for Robotic Transfer of Dexterous Human Manipulation
A hand-exoskeleton system that lets robots learn dexterous manipulation in-the-wild — combining tactile sensing with haptic feedback so demonstrations transfer cleanly from human hand to gripper.
Keypoint Abstraction using Large Models for Object-Relative Imitation Learning
KALM uses VLMs to automatically generate task-relevant keypoints, yielding stronger generalization for object-relative robot manipulation.
Diffusion Imitation from Observation
DIFO integrates diffusion models to model state transitions and supply robust rewards — improving policy learning from observation-only demonstrations, without action labels.
Improving XCO2 Precision in OCO-2/3 Retrievals through Machine-Learning Extraction of Volcanic Aerosol Information from L1B Spectra
A machine-learning pipeline that refines aerosol information from OCO satellite spectra, sharpening CO₂ retrieval accuracy and advancing climate models of volcanic aerosol impact.
// 06
service_and_teaching
Reviewing
- 2026 ICLRICRACoRLNeurIPSIROS
- 2025 NeurIPSICLR
- 2024 CoRL LangRob WS
Talks
- 2026 RL with Human FeedbackTutorial · AIxHRI Summer School, Nancy
- 2024 Diffusion Imitation from ObservationPoster · NeurIPS, Vancouver
Teaching
- F26 Robot LearningUSC CSCI 642 · TA
- F23 Reinforcement LearningNTU CommE5069 · TA
- S23 Signals and SystemsNTU EE2011 · TA
- S23–24 Cornerstone EECS DesignNTU EE1006 · TA