April 15-18, 2009
Indianapolis, Indiana, USA

Demonstrations: Description

A Responsive Virtual Human Museum Guide at the Boston Museum of Science

Timothy Bickmore, Northeastern University, USA

“Tinker” is a virtual museum guide installed at the entrance to the Computer Place exhibit at the Boston Museum of Science. Tinker appears as a six-foot-tall 3D cartoon robot, projected in front of visitors, and communicates with them using synthetic speech and synchronized nonverbal conversational behavior. A motion sensor causes Tinker to invite visitors who are walking by to talk to her, and a presence sensor triggers the system to begin and end conversations. Tinker provides visitors with information on and directions to a range of exhibits in the museum, and discusses the theory and implementation underlying its own creation. Technical content is tailored to each visitor’s level of computer literacy, assessed through dialogue. The dialogue system comprises 540 dialogue states.

Tinker is unique in that she uses human relationship-building behaviors to engage museum visitors. Tinker uses empathy, social dialogue, reciprocal self-disclosure and other relational behavior to establish social bonds with users. She also uses a biometric identification system so that she can re-identify visitors she has already talked to, and maintains persistent discourse and relational models, so that prior conversations and relationships can be seamlessly continued when visitors return.

Tinker was developed over an eight-month period in a collaboration between Northeastern University and the staff at Computer Place. This is a staffed area of the museum that provides visitors with explorations in computer science, communications, and robotics. Work on Tinker’s software infrastructure, dialogue content, character animation, and physical installation proceeded in parallel. Dialogue content was authored using a visual dialogue design tool developed to enable rapid construction of virtual humans. Tinker’s nonverbal conversational behavior is automatically generated based on linguistic analysis of its utterances and communication rules gleaned from studies of human-human interactions.

Pilot testing indicated that most visitors thought Tinker was fun and engaging to interact with, many (56%) preferred to talk to her rather than museum staff (only 31% said they would rather have talked to a person), and none expressed privacy concerns regarding the biometric identification system (78% had no concerns, the rest were unsure). Since her launch in April, 2008, over 30,000 visitors have interacted with Tinker. Most (94%) visitors conduct a single conversation lasting 7 minutes on average, with 6% returning for follow up conversations. The most popular topics that visitors ask about are Tinker’s design (41%), the Computer Place exhibit (23%), and directions to other parts of the museum (21%). An automated survey taken by 512 visitors indicated that most visitors are satisfied with the exhibit (1.81 on a scale of 1 for “very satisfied” to 5 for “not at all satisfied”) and like Tinker (1.91 on a scale of 1 for “very much” to 5 for “not at all”). Tinker continues to be used as a research test bed for automated guides that provide persistent, tailored, and personalized sources of information for museum visitors.

In this demonstration, I will show Tinker’s dialogue and animation systems, using simulated sensor input.

Demonstration: Demonstrations 2 [Close Up]

Keywords: relational agent, biometric visitor identification, animated character, simulated conversation