Kush Bhatia

University of California, Berkeley, Berkeley College

Papers

Total Citations

101

H-Index

About

Kush Bhatia’s research lies at the intersection of human-robot interaction, explainable AI, and trust calibration. His central contribution is developing methods to help non-expert users build accurate mental models of autonomous systems—particularly those driven by opaque neural network policies. His most influential work, “Establishing Appropriate Trust via Critical States” (2018), with 93 citations, introduces a novel approach that identifies and presents critical decision points to users, enabling them to grasp a robot’s capabilities and limitations without needing technical expertise. This work directly addresses the challenge of over- or under-trust in automation, a key barrier to safe human-robot collaboration. In his follow-up work, “Explaining Robot Policies” (2021), Bhatia extends this framework by systematically selecting examples that reveal a robot’s behavior across diverse situations, further refining how explanations can be tailored to improve user understanding. His research is notable for its practical, user-centered focus—moving beyond abstract interpretability to measurable improvements in human decision-making and trust. With a growing citation impact, Bhatia’s work is shaping how we design robots that are not only capable but also transparent and trustworthy partners in real-world settings.