Action understanding as a social-cognitive benchmark for human-aligned robotics

Chaolan Lin

Type: An evaluation-centric paper (focused on advancing methods and ideas for robot evaluation)

Keywords: Action understanding, human-aligned robotics, cognitive science, human-robot interaction, trust and transparency in robots

Abstract: As robots increasingly operate in everyday human environments, their success depends not only on task completion but also on how their actions are perceived and understood. This paper introduces an action understanding framework as a social-cognitive benchmark for evaluating whether robot behavior supports interpretability, transparency, and trust. We identify ten psychologically grounded action types and propose benchmark tasks that reflect core human cognitive mechanisms. By prioritizing cognitive legibility over performance metrics, the framework provides a principled and inclusive basis for designing and evaluating robots that are socially aligned, interpretable, and trustworthy.

Submission Number: 12