Justin Yu

University of California, Berkeley, Berkeley College

Papers

Total Citations

H-Index

About

Justin Yu’s research lies at the intersection of robotics and reinforcement learning, with a focus on making autonomous systems practical, sample-efficient, and deployable beyond controlled labs. His most influential work, “The Ingredients of Real-World Robotic Reinforcement Learning” (28 citations), systematically dissects the critical components—from hardware design to algorithmic choices—that enable robots to learn continuously in messy, real-world environments, moving beyond instrumented laboratory setups. This paper has become a foundational reference for researchers tackling the gap between simulation and physical deployment. Yu further advances exploration in reinforcement learning with “MURAL: Meta-Learning Uncertainty-Aware Rewards for Outcome-Driven Reinforcement Learning” (6 citations), where he introduces a meta-learning framework that leverages successful outcome examples to guide agents toward high-reward states more efficiently. This work addresses the notoriously difficult exploration problem by defining a tractable subclass of RL tasks. Through these contributions, Yu is shaping how roboticists think about data efficiency, reward design, and the practical ingredients necessary for robust, real-world learning systems—a critical step toward autonomous robots that can operate safely and adaptively outside the lab.