Junhong Xu

Indiana University, Indiana University Bloomington

Papers

Total Citations

H-Index

About

Junhong Xu is a researcher advancing the frontiers of data-efficient and causally-aware robotic learning. His work centers on three key areas: model-based reinforcement learning, causal inference for robotics, and multi-agent decision-making under bounded rationality. Xu’s major contribution includes a novel data-efficient reinforcement learning method that leverages local Koopman operators to drastically reduce the trial data required for training, addressing a fundamental bottleneck in applying RL to real-world systems. He further tackles the pervasive problem of biased observational data in robotics by introducing a principled causal inference framework that de-biases motion estimation, enabling more reliable learning from real-world deployments. His recent work extends to decision-making among bounded rational agents, exploring how limited computational resources affect strategic interactions. While his publication record is still growing, with his most-cited paper garnering 5 citations, the conceptual rigor and practical relevance of his research—particularly in bridging model-based RL and causal reasoning—signal a promising trajectory. Xu’s contributions are especially notable for their potential to make robot learning safer, more efficient, and more robust in complex, real-world environments.