This thesis introduces a novel behavior categorization model that can be used for behavior recognition and learning. Correlation Based Adaptive Resonance Theory (CobART) network, which is a kind of self organizing and unsupervised competitive neural network, is developed for this purpose. CobART uses correlation analysis methods for category matching. It has modular and simple architecture. It can be adapted to different categorization tasks by changing the correlation analysis methods used when needed. CobART networks are integrated hierarchically for an adequate categorization of behaviors. The hierarchical model is developed by adding a second layer CobART network on top of first layer networks. The first layer CobART networks categorize self behavior data of a robot or an object in the environment. The second layer CobART network receives first layer CobART network categories as an input, and categorizes them to elicit the robot&#039 / s behavior with respect to its effect on the object. Besides, the second layer network back-propagates the matching information to the first layer networks in order to find the relation between the first layer categories. The performance of the hierarchical model is compared with that of different neural network based models. Experiments show that the proposed model generates reasonable categorization of behaviors being tested. Moreover, it can learn different forms of the behaviors, and it can detect the relations between them. In essence, the model has an expandable architecture and it contains reusable parts. The first layer CobART networks can be integrated with other CobART networks for another categorization task. Hence, the model presents a way to reveal all behaviors performed by the robot at the same time.

Communication in HRI, both verbal and non-verbal, can be hard for a robot to interpret and to convey which can lead to misinterpretations by both the human and the robot. In this thesis we look at answering the question if AR can be used to improve communication of a social robot’s intentions when interacting with children. We looked at behaviors such as getting children to pick up a cube, place a cube, give the cube to another child, tap the cube and shake the cube. We found that picking the cube was the most successful and reliable behavior and that most behaviors were slightly better with AR. Additionally, endorsement behavior was found to be necessary to engage the children, however, it needs to be quicker, more responsive and clearer. In conclusion, there is potential for using AR to improve the intent communication of a robot, but in many cases, the robot behavior alone was already quite clear. A larger study would need to be conducted to further explore this. / I Människa-Robot Interaktion kan både verbal och icke-verbal kommunikation vara svårt för en robot att förstå och förmedla vilket kan leda till missförstånd från både människans och robotens håll. I den här rapporten vill vi svara på frågan ifall AR kan användas för att förbättra kommunikationen av en social robots avsikter när den interagerar med barn. De beteenden vi kollade på var att få ett barn att plocka upp en kub, placera den, ge den till ett annat barn, knacka på den och skaka den. Resultaten var att plocka upp kuben var det mest framgångsrika och pålitliga beteendet och att de flesta beteenden var marginellt bättre med AR. Utöver det hittade vi också att bifallsbeteenden behövdes för att engagera barnen men behövde vara snabbare, mer responsiva och tydligare. Sammanfattningsvis finns det potential för att använda AR, men i många fall var enbart robotens beteenden redan väldigt tydliga. En större studie skulle behövas för att utforska detta ytterligare.

In this thesis, behavior learning and generation models are proposed for simple and complex behaviors of robots using unsupervised learning methods. Simple behaviors are modeled by simple-behavior learning model (SBLM) and complex behaviors are modeled by complex-behavior learning model (CBLM) which uses previously learned simple or complex behaviors. Both models have common phases named behavior categorization, behavior modeling, and behavior generation. Sensory data are categorized using correlation based adaptive resonance theory network that generates motion primitives corresponding to robot&#039 / s base abilities in the categorization phase. In the modeling phase, Behavior-HMM, a modified version of hidden Markov model, is used to model the relationships among the motion primitives in a finite state stochastic network. In addition, a motion generator which is an artificial neural network is trained for each motion primitive to learn essential robot motor commands. In the generation phase, desired task is presented as a target observation and the model generates corresponding motion primitive sequence. Then, these motion primitives are executed successively by the motion generators which are specifically trained for the corresponding motion primitives. The models are not proposed for one specific behavior, but are intended to be bases for all behaviors. CBLM enhances learning capabilities by integrating previously learned behaviors hierarchically. Hence, new behaviors can take advantage of already discovered behaviors. The proposed models are tested on a robot simulator and the experiments showed that simple and complex-behavior learning models can generate requested behaviors effectively.