Human-Robot Plan Communication

Abstract

Although the technical development of robots has made them more autonomous, they still benefit from some human skills and advice. Therefore, the robot needs to interact with a human and ask him for help to get out of difficulties in the best way. This dissertation addresses the problem of finding better mechanisms to communicate to a robot the directions for navigation in indoor environments. We identify which out of a set of combinations of speech, gestures, and drawing mechanisms are the most comfortable, easier to learn, and least error-prone for human users. Three different methods: a Speaking method, a Gesturing and Speaking method, and a Drawing method were tested for guiding the NAO robot to the desired place, assuming that the robot has no prior map of the building and even no information about the involved human faces. Our experiment consists of having participants ask the robot to accomplish a complex indoor navigation task. Task communication is attempted with each of the above methods. In one experiment, no additional details were provided about the proposed methods to help the participants, except for a video showing the problem in neuter human terms. In subsequent tests, the participants were also given restrictions concerning the robot communication capabilities: only to use speech, gestures, or vision. In the third set of tests, the participants were also given training using examples of efficient communication with corresponding methods. The task was decomposed, with points assigned for each component. The methods have been investigated and evaluated based on the average task success in simulated plan execution. In our results, the Drawing method leads to the highest level of task accomplishment, and it is also the fastest communication method. Further, according to the final short questionnaire, most participants feel that the Drawing method is more comfortable and they can use it immediately without thinking, as compared to methods based on speech.