The paper discusses the challenges of engineering applications for humanoid robots like Pepper. Such robots can be used in di erent domains. We focus on the training aspect for patients after a stroke. The humanoid plays the role of a trainer. It should motivate the patient but also pushes when necessary. It should recognize the need for brakes and provide social interactions. A speci c challenge is the collaboration of a patient, a supporting person and the robot. This collaboration should be modeled before implementation. Additionally, a model of the three roles is necessary. Based on the analyzed situation an appropriate interaction has to be selected. Domain speci c languages might be helpful for modeling and engineering the applications.
Robots are used in di erent domains. They are used to support industrial
production [
Pepper can be programmed in Java with the QiSDK plug-in for Android Studio. Fig. 2 provides a screenshot of the programming environment with the plugin extensions for the robot. The plugins allow the connection with a real or a virtual robot. Additionally, on can see the virtual simulator of Pepper. It looks like the real robot and has arms and ngers but wheels instead of legs.
In our project we want to study whether robots like Pepper are able to motivate patients for their exercises. Currently, exercises are observed by humans and can be provided only two or three times per week. However, it would be much better if patients would be able to train several times per day.
Unfortunately, success results of the training come very slowly. Even that there is some success after a week, patients often do not recognize it. That is the reason, why they often question the sense of their exercises. Therefore, motivation is very important. We assume that we can support the motivation of a patient with the help of the robot Pepper. One option is the expression of the observed success by Pepper in an appropriate way.
Together with psychologists, sociologists and experts from medicine we want to study which kind of patients can be supported by Pepper in which way. We want to provide answers for questions like: What kind of interaction is appropriate in certain circumstances? When does a patient need a support? When does it make sense to suggest a break? How can motivation be increased? What kind of models can be helpful? How can we engineer applications for Pepper in an appropriate way?
Our project was initiated by Professor Thomas Platz who has been working on
Neurohabilitation for several years. He invented some speci c exercises for arms
that stimulate the brain of stroke patients. An evaluation of his approach is
published in [
In the left upper corner of Fig. 3 one can see how a patient has to hit di erent circles. The time for this exercise is two minutes. A patient has to hit the two largest circles rst, then the two smaller ones and so on. After hitting the smallest one the patient has to start from the beginning again. Currently, a physiotherapist counts the correct hits. If this training is supported by a tablet implementation one can count the hits and their accuracy more precisely. The robot Pepper can comment on the reached results and motivate a patient with sentences like: Great, you reached a new level! Let us try again, we nearly reached the best result! Today, you are really in a good mood. Let us make a break and try to reach the same result again!
Some patients have at the beginning such problems with their arm that they are not able to hold it by themselves. They need a supporter that provides guidance by holding the elbow of the patient in its hand. This can be boring for the supporter after a while. Therefore, both the patient and the supporter need some motivation for the training. Pepper has to recognize if the motivation of one of the collaborators drops. If the motivation level of the patient and the supporter falls at the same time it has to be decided with whom to interact rst. This decision has to be made based on the expected cognitive state of the collaborators. Knowledge representations have to be speci ed for such application.
Research has been focused during the last years on the interaction of a
computer with one user. Two users and a greater variety of interaction modes provide
new challenges. There are only few papers about Robot interaction with several
people. Tahir et al. [
As a rst attempt we will try to use our language DSL-CoTaL [
Fig. 4 provides an impression how the language looks like. The behavior of
Pepper is speci ed as a task model in a rule-based way. A task on the left hand
side of a rule is speci ed by sub-tasks on the right side. Some tasks need a
precondition before they can be executed. Preconditions are expressed in an OCL-like[
The task model of the patient and of the supporter are omitted here. They
do not provide new insights. However, a simple collaboration model can be seen
in Fig. 5. It is written in the style of the collaboration model of CTTE [
Especially for the activities of the robot Pepper it makes sense to specify a
new domain-speci c language (DSL) that allows the code generation for the
Android Studio. Tasks like raise left arm would become keywords in that language.
In conjunction with the project E-BRAiN we identi ed the following research
questions:
{ How can applications with social robots like Pepper be engineered?
{ What kind of interaction is appropriate for which kind of patient or supporter
in which context?
{ How can a concept be developed for systems with two persona types in
connection with two user models? Which ideas can be reused from two-level
personas [
These questions (and de nitely some more) will guide us in our project during the next three years. 4
The challenges in engineering of social humanoid robots like pepper were discussed. In context of the project E-BRAiN collaborative systems involving a robot, a patient and a supporter have to be developed. Managing the appropriate interaction technology (voice, gesture, lights, tablet, other devices) is a challenge. Domain-speci c languages seem to be one way to specify necessary models. Some ideas were discussed and some research questions were presented. Further ideas are expected from the discussion during the workshop.