Jingkang Wang

Papers

Total Citations

133

H-Index

About

Jingkang Wang is a researcher whose work spans reinforcement learning robustness and realistic sensor simulation, with contributions that have meaningfully advanced both theoretical foundations and practical applications in machine learning and robotics. His most influential work, "Reinforcement Learning with Perturbed Rewards," has accumulated over 100 citations since its publication, addressing a critical vulnerability in RL systems: the unreliability of reward signals collected through noisy channels such as physical sensors. By developing principled methods for training robust RL agents under reward perturbation, Wang helped lay groundwork for deploying reinforcement learning in real-world, imperfect environments — a problem of considerable practical importance. His more recent contribution, NeuSim, tackles a different but equally challenging frontier: reconstructing real-world objects from sparse data to enable high-fidelity sensor simulation for robotics training and testing. This work reflects a broadening of his research agenda toward scalable, realistic simulation pipelines that reduce dependency on costly real-world data collection. Together, these contributions position Wang as a researcher thoughtfully bridging gaps between theoretical machine learning and the messy demands of real-world robotic and autonomous systems deployment.