Candidate: Pourya Aliasghari
Date: July 15, 2024
Time: 12:00 PM
Place: E5 5047
Supervisor(s): Dautenhahn, Kerstin - Nehaniv, Chrystopher L. - Ghafurian, Moojan
All are welcome!
Abstract:
For social robots to succeed in places such as homes, they must learn new skills from various people and act in a manner desirable to different users. Our aim is to enable robots to learn from people who may have different individual preferences, in a manner that is familiar and natural to us, i.e., through social learning. Sometimes, users may not know how to teach new tasks to a robot. In our first step, we studied whether participants without any experience in teaching a robot would become more proficient robot teachers through repeated kinesthetic human-robot teaching interactions. An experiment was conducted with twenty-eight participants who were asked to teach a humanoid robot different cleaning tasks in multiple sessions. Our data analyses revealed a diversity in non-experts' human-robot teaching styles in repeated interactions. The majority of participants significantly improved their success and speed of kinesthetic demonstrations by performing multiple rounds of teaching the robot. In the next phase, we introduce a novel biologically inspired approach for robot learning through program-level imitation, inspired by the way primates, including humans, understand and perform complex actions. Our approach enables robots to discover the hierarchical structure of tasks by identifying sequential regularities and sub-goals from diverse human demonstrations. We implemented our system on an iCub humanoid robot and present how our method allows the robot to adapt its action sequences for task execution, taking into account user preferences. This system is meant to accommodate variations in human teaching styles and is expected to help a robot perform tasks with greater flexibility and efficiency and learn from humans at an abstract level, opening the way to more adaptable and intelligent robotic assistants in everyday tasks.