Minjune Hwang. Personal Webpage

Minjune Hwang

minjuneh at usc dot edu

Hello! I am a first year PhD Student at USC, where I am advised by Prof. Daniel Seita and closely working with Prof. Erdem Biyik.

I received a MSCS degree at Stanford, where I worked on robot learning research with Prof. Fei-Fei Li and Prof. Jiajun Wu. Prior to that I completed my undergraduate studies at UC Berkeley, majoring in CS & Statistics.

Previously, I was an Applied Scientist Intern at Amazon Robotics's Scanless Tech, where I developed an efficient algorithm that eliminates the need for explicit scanning, with Dr. Frank Preiswerk and Preethi Krishnaswamy. I have also interned in Microsoft Research and Apple (SPG) to solve various problems in robotics.

Resume / GitHub / LinkedIn / Google Scholar

Research (All / Highlighted )
(* indicates equal contribution, ^† indicates equal advising)

I am interested in developing robotic systems that can learn from human feedback and interact with humans. Specifically, my research focuses on desiging algorithms that can robustly and effectively reflect users' true preference and finetuning large pretrained policies with various kinds of human feedback to continuously adapt to their needs.

	Causally Robust Preference Learning with Reasons Minjune Hwang, Yigit Korkmaz, Daniel Seita^†, Erdem Biyik^† HiTL Workshop @ RSS 2025 (Oral) paper \| poster PbRL is widely used for shaping agent behavior to match a user's preference, yet its sparse binary feedback makes it vulnerable to causal confusion. We introduce ReCouPLe, a lightweight framework that uses natural language rationales to clarify true causal signals behind preference and to improve generalization, by employing orthogonal decomposition.
	Decentralized Vehicle Coordination: The Berkeley DeepDrive Drone Dataset and Consensus-Based Models Fangyu Wu, Dequan Wang, Minjune Hwang, Chenhui Hao, Jiawei Lu, Jiamu Zhang, Christopher Chou, Trevor Darrell, Alexandre Bayen ICRA 2025 PAL Workshop @ ICRA 2020 (Oral) arXiv \| code We introduce a novel traffic dataset and modeling framework tailored to study motion planning in understructured environments. In this project, I was in charge of the object detection and tracking framework for constructing the dataset.
	NOIR: Neural Signal Operated Intelligent Robots for Everyday Activities Ruohan Zhang, Sharon Lee, Minjune Hwang, Ayano Hiranaka, Chen Wang, Wensi Ai, Jin Jie Ryan Tan, Shreya Gupta, Yilun Hao, Gabrael Levine, Ruohan Gao, Anthony Norcia, Li Fei-Fei, Jiajun Wu CoRL 2023 CRL Workshop @ CoRL 2023 (Oral) paper \| project page NOIR is a general-purpose BRI system that enables humans to command robots to perform everyday activities with their brain signals. Here, I designed a hierarchical EEG decoding system with parameterized skills, as well as a few-shot imitation learning algorithm.
	Primitive Skill-Based Robot Learning from Human Evaluative Feedback Minjune Hwang, Ayano Hiranaka, Sharon Lee, Chen Wang, Li Fei-Fei, Jiajun Wu, Ruohan Zhang IROS 2023 arXiv \| project page \| poster SEED is an RLHF algorithm that leverages primitive skills to enable more safe and sample efficient long-horizon task learning. I mainly designed the hierarchical RLHF algorithm and led simulation experiments.
	Task-Driven Graph Attention for Hierarchical Relational Object Navigation Michael Lingelbach, Chengshu Li, Minjune Hwang, Andrey Kurenkov, Alan Lou, Roberto Martin-Martin, Ruohan Zhang, Li Fei-Fei, Jiajun Wu ICRA 2023 arXiv \| github We propose a novel solution to object navigation tasks based on a scene graph representation, that combines graph neural networks and task-driven attention for better scalability and learning efficiency. I empirically designed the task-driven graph attention and verified its efficacy in simulation environments.
	BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation Chengshu Li, Cem Gokmen, Gabrael Levine, Roberto Martin-Martin, Sanjana Srivastava, Chen Wang, Josiah Wong, Ruohan Zhang, Michael Lingelbach, Jiankai Sun, Mona Anvari, Minjune Hwang, Manasi Sharma, Arman Aydin, Dhruva Bansal, Samuel Hunter, Kyu-Young Kim, Alan Lou, Caleb R Matthews, Ivan Villa-Renteria, Jerry Huayang Tang, Claire Tang, Fei Xia, Silvio Savarese, Hyowon Gweon, Karen Liu, Jiajun Wu, Li Fei-Fei CoRL 2022 (Best Paper Nomination; Oral) arXiv \| project page \| github BEHAVIOR-1K is a benchmark for embodied AI and robotics research with realistic simulation of 1,000 diverse household activities grounded in human needs. As a core developer, I developed generalizable robot kinematic modules and controllers, as well as a foundational library of primitive skills for mobile manipulation.
	Minority Reports Defense: Defending Against Adversarial Patches Michael McCoyd, Won Park, Steven Chen, Neil Shah, Ryan Roggenkemper, Minjune Hwang, Jason Xinyu Liu, David Wagner ACNS 2020 SiMLA @ ACNS 2020 (Best Paper Award) arXiv \| github We propose a defense against patch attacks based on partially occluding the image around each candidate patch location, so that a few occlusions each completely hide the patch. I mainly developed the sparsity-invariant CNNs for adversarial attack detection via occlusion.
	Text Analytics for Resilience-Enabled Extreme Events Reconnaissance Alicia Yi-Ting Tsai, Selim Gunay, Minjune Hwang, Chenglong Li, Pengyuan Zhai, Laurent El Ghaoui, Khalid M.Mosalam AI + HADR @ NeurIPS 2020* (Spotlight) arXiv \| project page \| github We develop a text analytics framework to extract information from social media data for disaster response. I mainly designed the text analytics framework which used sparse convex optimization and NLP methods.

Teaching

Graduate Teaching Assistant: Stanford CS 231N [2023], Deep Learning for Computer Vision

Reader (Undergraduate Teaching Assistant): UC Berkeley EE 227BT [2020], Convex Optimization

Course Instructor: Ecole Bilingue de Berkeley [2019], Robotics & Programming (with Prof. Alex Bayen)

Undergraduate Lab Assistant: UC Berkeley CS 61A [2018], Structure and Interpretation of Computer Programs

Industry Experience

Applied Scientist Intern, Amazon Robotics

Research Intern, Microsoft

Software Engineering Intern (Motion & Trajectory Planning), Apple (SPG)

Honors & Fellowship

Viterbi School of Engineering Fellowship Aug. 2024 - Jul. 2026

High Distinction in General Scholarship (Magna Cum Laude) May. 2021

Summer Undergraduate Research Fellowship (SURF), UC Berkeley May. 2020

Berkeley Undergraduate Scholarship Aug. 2017 - May. 2021

Service

Serving/Served as a reviewer for CoRL & IROS, as well as workshops in RSS.

Serving as a PhD mentor for a number of undegraduate students and summer interns at USC

Inspired by this and this.

	Causally Robust Preference Learning with Reasons Minjune Hwang, Yigit Korkmaz, Daniel Seita^†, Erdem Biyik^† HiTL Workshop @ RSS 2025 (Oral) paper \| poster PbRL is widely used for shaping agent behavior to match a user's preference, yet its sparse binary feedback makes it vulnerable to causal confusion. We introduce ReCouPLe, a lightweight framework that uses natural language rationales to clarify true causal signals behind preference and to improve generalization, by employing orthogonal decomposition.
	Decentralized Vehicle Coordination: The Berkeley DeepDrive Drone Dataset and Consensus-Based Models Fangyu Wu, Dequan Wang, Minjune Hwang, Chenhui Hao, Jiawei Lu, Jiamu Zhang, Christopher Chou, Trevor Darrell, Alexandre Bayen ICRA 2025 PAL Workshop @ ICRA 2020 (Oral) arXiv \| code We introduce a novel traffic dataset and modeling framework tailored to study motion planning in understructured environments. In this project, I was in charge of the object detection and tracking framework for constructing the dataset.
	NOIR: Neural Signal Operated Intelligent Robots for Everyday Activities Ruohan Zhang, Sharon Lee, Minjune Hwang, Ayano Hiranaka, Chen Wang, Wensi Ai, Jin Jie Ryan Tan, Shreya Gupta, Yilun Hao, Gabrael Levine, Ruohan Gao, Anthony Norcia, Li Fei-Fei, Jiajun Wu CoRL 2023 CRL Workshop @ CoRL 2023 (Oral) paper \| project page NOIR is a general-purpose BRI system that enables humans to command robots to perform everyday activities with their brain signals. Here, I designed a hierarchical EEG decoding system with parameterized skills, as well as a few-shot imitation learning algorithm.
	Primitive Skill-Based Robot Learning from Human Evaluative Feedback Minjune Hwang, Ayano Hiranaka, Sharon Lee, Chen Wang, Li Fei-Fei, Jiajun Wu, Ruohan Zhang IROS 2023 arXiv \| project page \| poster SEED is an RLHF algorithm that leverages primitive skills to enable more safe and sample efficient long-horizon task learning. I mainly designed the hierarchical RLHF algorithm and led simulation experiments.
	Task-Driven Graph Attention for Hierarchical Relational Object Navigation Michael Lingelbach, Chengshu Li, Minjune Hwang, Andrey Kurenkov, Alan Lou, Roberto Martin-Martin, Ruohan Zhang, Li Fei-Fei, Jiajun Wu ICRA 2023 arXiv \| github We propose a novel solution to object navigation tasks based on a scene graph representation, that combines graph neural networks and task-driven attention for better scalability and learning efficiency. I empirically designed the task-driven graph attention and verified its efficacy in simulation environments.
	BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation Chengshu Li, Cem Gokmen, Gabrael Levine, Roberto Martin-Martin, Sanjana Srivastava, Chen Wang, Josiah Wong, Ruohan Zhang, Michael Lingelbach, Jiankai Sun, Mona Anvari, Minjune Hwang, Manasi Sharma, Arman Aydin, Dhruva Bansal, Samuel Hunter, Kyu-Young Kim, Alan Lou, Caleb R Matthews, Ivan Villa-Renteria, Jerry Huayang Tang, Claire Tang, Fei Xia, Silvio Savarese, Hyowon Gweon, Karen Liu, Jiajun Wu, Li Fei-Fei CoRL 2022 (Best Paper Nomination; Oral) arXiv \| project page \| github BEHAVIOR-1K is a benchmark for embodied AI and robotics research with realistic simulation of 1,000 diverse household activities grounded in human needs. As a core developer, I developed generalizable robot kinematic modules and controllers, as well as a foundational library of primitive skills for mobile manipulation.
	Minority Reports Defense: Defending Against Adversarial Patches Michael McCoyd, Won Park, Steven Chen, Neil Shah, Ryan Roggenkemper, Minjune Hwang, Jason Xinyu Liu, David Wagner ACNS 2020 SiMLA @ ACNS 2020 (Best Paper Award) arXiv \| github We propose a defense against patch attacks based on partially occluding the image around each candidate patch location, so that a few occlusions each completely hide the patch. I mainly developed the sparsity-invariant CNNs for adversarial attack detection via occlusion.
	Text Analytics for Resilience-Enabled Extreme Events Reconnaissance Alicia Yi-Ting Tsai, Selim Gunay, Minjune Hwang, Chenglong Li, Pengyuan Zhai, Laurent El Ghaoui, Khalid M.Mosalam AI + HADR @ NeurIPS 2020* (Spotlight) arXiv \| project page \| github We develop a text analytics framework to extract information from social media data for disaster response. I mainly designed the text analytics framework which used sparse convex optimization and NLP methods.