We propose a knowledge representation of everyday objects based on prototype theory in order to improve human-robot interaction. Since robots are becoming increasingly important in our everyday life, one day they might be used to do the chores, e.g. in kitchens. In order to tidy objects, however, robots have to be able to find the places where items belong; thus they need to categorize them. We develop a paradigm that mimics human categorization, in order to provide flexible, human-like solutions.
In order to identify a suitable approach to the prototype theory, we augmented and implemented two approaches, and compared their performance. We found that prototype models represent similarities between objects well. Further, we found out that one approach (Minda and Smith, 2011) is preferable over the other (Hampton, 1993), although on the whole they do not differ too much.
We implemented and augmented the approach described by Minda and Smith in order to contribute to a human-like knowledge representation.
We let people sort objects in a simulated as well as in a real kitchen to research human’s applied grouping behaviour. We compared these human clusters to those found by artificial clustering methods and found out, that artificial clustering can be done more human-like if features are weighted in a human way.
In another experiment we let participants group kitchen objects without a kitchen. One research question of this study is whether artificial clustering can “learn” from human clustering. Another question is concerning the influence of the kitchen on the grouping.
We also let participants group Lego bricks in no more than six groups. Then we categorized new test bricks by humans, our algorithms and two alternative algorithms into these six groups. One last experiment showed that participants do not need significant more time and trials to find the test bricks when they were categorized by our algorithm than when they were categorized by humans.
One final question is to what extent the proposed knowledge representation is able to reflect human categorizations in a kitchen and if the resulting behaviour of a robot is intuitively understandable for users in this case as well.
Doctoral Proposal of Alisa Volkert.
