Out of the Loop: Bounded LLM Mediation for Cyber-Physical Control
Under review, NeurIPS 2026
A dual-agent LLM mediation architecture that coordinates cyber-physical agents through bounded action masking and reward shaping.
Hi! I am a first-year Ph.D. student in Civil Engineering at the University of Southern California, advised by Prof. Ruolin Li. My research lies at the intersection of multi-agent learning, game theory, and LLM-based decision making for cyber-physical systems.
I am especially excited about macro-micro agent systems: how a system-level coordinator can shape many local agents, and how local agent behaviors feed back into global dynamics. My current projects include bounded LLM mediation for cyber-physical control, defensive population learning for competitive autonomous ride-hailing, and macro-micro reinforcement learning for mixed-autonomy mobility.
Currently thinking about: macro ↔ micro agents LLM-mediated control multi-agent RL game-theoretic learning
Email: kwang255@usc.edu
News
- Oct 2026 Heading to INFORMS Annual Meeting in San Francisco for an oral presentation.
- Jul 2026 Next stop: INFORMS Transportation Science and Logistics Conference at MIT.
- Apr 2026 Shared recent work at the SoCal CEE Research Symposium at USC.
- Jan 2026 Presented at the Transportation Research Board Annual Meeting in Washington, D.C.
- Oct 2025 Gave an oral presentation at INFORMS Annual Meeting in Atlanta.
- Sep 2025 Presented a poster at the USC Center for Autonomy & AI Workshop.
- Jul 2025 Gave an oral presentation at the USC STEM Bytes Seminar.
- Apr 2025 Presented at the SoCal CEE Research Symposium at UCI.
Publications & Under Review
Beyond Meta-Nash: Defensive PSRO for Robust Learning under Non-Transitive Competition
In preparation, AAAI 2026
A defensive population-based learning framework for robust decision-making against off-equilibrium competitive strategies.
Macro-Micro Pigouvian Multi-Agent Reinforcement Learning for Mixed-Autonomy Lane Control
Under review, IEEE Conference on Decision and Control (CDC 2026)
A macro-micro MARL framework where lane-level Pigouvian signals coordinate decentralized CAV agents.
When Altruism Meets Autonomy: Managing Weaving Ramp Congestion with Strategic AVs
Under review, Transportation Science
A Stackelberg-Wardrop framework for strategic autonomous agents shaping equilibrium behavior.
Traffic Equilibrium in Mixed-Autonomy Network with Capped Customer Waiting
Under review, Transportation Research Part B
A network equilibrium model for mixed AV/HV ride-hailing fleets, customer waiting, and platform decisions.
To Stay or to Bypass: Unraveling Mainline Vehicles' Aggregate Strategic Decision-Making in Weaving Ramps
IEEE ITSC 2025
A Wardrop-equilibrium formulation of aggregate lane-choice behavior for mainline vehicles at highway weaving ramps.
Teaching & Mentoring
- Teaching Assistant, CE-119 Statistical Data Analysis in Engineering, USC, Fall 2025.
- Ph.D. Mentor, USC CURVE Program, mentoring undergraduate research projects across CS, EE, and CEE, 2025-2026.
Academic Services
- Organizer, 2026 Conference on Robot Learning (CoRL) workshop.
- Website Designer, 2026 IFAC Workshop on Cyber-Physical Human Systems (CPHS).
- Student Volunteer, 2026 NASA Formal Methods Symposium (NFM).
- Reviewer, BTR 2026, IEEE CDC 2025/2026, IEEE ITSC 2026, IEEE IV 2026, IEEE ICRA 2026, TRB 2025.