Wen-Tse Chen 陳文澤

I am a second-year student in the Master of Robotics program at Carnegie Mellon University advised by Prof. Jeff Schneider. Prior to this, I pursued my undergraduate studies in Automation at Tsinghua University, where I had the privilege of working alongside Prof. Jun Zhu. I also gained research experience under the guidance of Prof. Koushil Sreenath at UC Berkeley.

I am currently applying to Ph.D. programs for Fall 2025.

Email  /  Google Scholar  /  Github

My research interests focus on LLM agents, multi-agent reinforcement learning, and their applications in decision-making and robotics.

Presented a sample-efficient method for online fine-tuning LLM agents by using in-context learning to convert sparse feedback into dense signals, enabling LLMs to adapt to dynamic environments with minimal data.

Enhanced QMIX with maximum entropy RL for better exploration, ensuring locally optimal actions match global optima via an order preserving transformation, achieving SOTA performance on SAMC-v2 benchmark.

An open source framework supports single-agent RL, multi-agent RL, RLHF, and self-play training.

Proposed an on-policy framework for discovering multiple diverse optimal strategies for the same task in a single training process.

Created an on-policy MARL algorithm, along with an adaptive curriculum learning approach and a unique self-play strategy, for excelling in the Google Research Football game.

Developed a distributed learning system and new offline algorithms to learn a powerful multi-agent AI from the fixed single-agent dataset.

last update: Oct 30th, 2024

Copy from Dr. Jon Barron's page.