This thesis studies mechanisms to improve human-robot-interaction through a spoken dialogue for household
robots. Therefore, a full dialogue system, in which the semantics of the words play an important
role, is implemented.
Nowadays, robots are found to be helpful in a lot of applications. One important eld during the
last decades is to design household robots capable of helping people with disabilities. For this purpose,
the robot has to be able to communicate with humans so it knows what it has to do. A natural way
to do so is by speech, which still needs further research. With the controller implemented, the robot is
expected to grasp objects, navigate, learn and follow people, clean up a room, etc. So the rst goal of
this thesis is to give the robot the ability to understand these tasks through a spoken dialogue system.
Furthermore, the robot has to be able to understand complex tasks where the information to achieve the
task is incomplete. The second goal of this thesis is that the robot has the ability to learn new words, for
example, names of people, because it is impossible to have all the names stored in a database in advance.
In addition, learning people not only implies learning the physical body, e.g., the face, but also their
names.
The dialogue system implemented has an accuracy of 80.45% for isolate words, 90.56% for simple
commands and 96.66% for complex commands that provide incomplete information for the robot to
achieve the task. Finally, the accuracy of recognizing words learnt on-line is 33.33%.
