Action Semantics

Action semantics research focuses on understanding and representing the meaning of actions, bridging the gap between observed motion or behavior and their underlying intentions and effects. Current efforts leverage large language models (LLMs) to infer action semantics from diverse data sources, including textual descriptions, visual observations, and even environmental feedback during robotic planning. This work aims to improve the robustness and generalizability of systems that reason about and predict actions, with applications ranging from autonomous navigation and human activity understanding to more efficient and explainable AI planning.