Unified Strategy for Fast Goal Inference in Human-Robot Interaction
A new method enhances goal inference in human-robot interactions through Critical Decision Points.
Researchers propose a novel approach for rapid goal inference in human-robot interactions by directing humans towards Critical Decision Points (CDPs). These CDPs are defined as states where different human strategies suggest varying actions, thereby clarifying the human's goal. The method integrates a goal-conditioned policy divergence measure into a Receding-Horizon Planner, which evaluates future actions while balancing task progress and information gain.
The approach was tested in both a collaborative cooking task and a competitive hide-and-seek game, demonstrating improved accuracy and speed in inferring human goals compared to traditional methods.