Start

End

KOÇ UNIVERSITY

GRADUATE SCHOOL OF SCIENCES & ENGINEERING

COMPUTER SCIENCES AND ENGINEERING

MS THESIS DEFENSE BY DOĞANCAN KEBÜDE

 

Title: Human-Inspired Communicative Cues for Intent Expressive Motion Generation

 

Speaker: Doğancan Kebüde

 

Time: 15 January 2019, 5:30 PM

 

Place: ENG 208

Koç University

Rumeli Feneri Yolu

Sariyer, Istanbul

Thesis Committee Members:

Asst. Prof. Barış Akgün (Advisor, Koç University)

Asst. Prof. Mehmet Gönen (Koç University)

Asst. Prof. Emre Uğur (Boğaziçi University)

Abstract: This thesis, inspired by the joint action literature, utilizes subtle, non-verbal kinematic changes that are not consciously recognizable in human motion called “communicative cues” to improve intent expressiveness of robot motion. Towards this end, this study first identifies two temporal and two spatial communicative cues through a human-human experiment (n = 14) and evaluates their utility in a simulated robot experiment (n = 30). After verifying the efficacy of communicative cues for generating intent expressive robot motion, an automated intent expressive motion generation framework is developed that handles the communicative cues through cue parameters. Via utilization of this motion generation framework, the communicative cues’ efficacy in real world human-robot interaction scenarios is verified and their robustness in different viewpoint and object configurations is evaluated in a human-robot experiment (n = 34). As the cue parameters are chosen manually in the human-robot experiment, an active reinforcement learning methodology is utilized to enhance the motion generation framework. Through a follow up human-robot learning experiment (n = 30), the proposed method is analyzed and it was seen that the robot can come up with user-specific cue parameters. Furthermore, a final human-robot experiment (n = 11) revealed that there might also be universal cue parameters that improve intent understanding for all users. This thesis contributes to the body of literature through investigation of human motion to identify the communicative cues, verification of the cues through a two-layer systematical analysis and evaluation of their robustness in different viewpoint and object configurations. Furthermore, an intent expressive motion generation framework is developed and it is enhanced with active reinforcement learning to achieve an improved intent expressive motion.

 

 

MORE DETAIL