Andi Peng

IIT@MIT, University of California, Berkeley

Papers

Total Citations

H-Index

About

Andi Peng is an emerging researcher whose work sits at the intersection of human-robot interaction, representation learning, and robot alignment. Her research addresses a fundamental challenge in modern robotics: ensuring that robots don't merely function effectively in isolation, but genuinely understand and align with human values, preferences, and intentions. Peng's most influential contribution, "Aligning Human and Robot Representations" (2024, 16 citations), tackles the critical gap between how robots internally model the world and how humans perceive meaningful task features. Rather than optimizing purely for functional efficiency, her work argues that robot representations must reflect human-relevant semantics — a subtle but transformative shift in how we design intelligent systems. This theme carries through her earlier 2023 version of the work, demonstrating a sustained commitment to this research direction. Her paper on "Preference-Conditioned Language-Guided Abstraction" (2024) further extends these ideas by leveraging natural language to construct more generalizable state representations for robot learning from demonstration, reducing problematic spurious correlations that plague imitation learning systems. Though early in her career, Peng's focus on human-centered alignment positions her at the forefront of a crucial conversation in robotics — one that will only grow more important as autonomous systems become deeply integrated into everyday human environments.