Real-time imperfect data impose a big challenge to safe driving of unmanned vehicles, where intelligent systems must learn potential dangers from complicated environments and unpredictable situations. This article aims to tackle the interaction of unmanned vehicles with human mind for self-controlling of the direction, speed, and posture. This interaction is explained under some necessary hypotheses and theoretically analyzed, where its mechanisms are interpreted as an integration of attention processes, thinking processes, and behavior processes. Attention processes are formulated by a locally squeeze-and-excitation network. The self-controlling behaviors of unmanned vehicles depend on intelligent decisions from deep thinking. A deep-thinking prototypical network (DTPNet) is introduced and used in perspective analyses, the interaction of unmanned vehicles with human mind. Results show that the perspective accuracies of mind interaction and self-controlling can approach to 93.1–93.8%. At the end of the article, the constraints of such interaction are also discussed.