Predicting 3D action targets from egocentric vision enhances human-robot interaction safety and efficiency.
