Graduate Research Talk: Coordination through Dialogue: Applications in Human-Robot Teaming
Abstract
Robots are increasingly needed to serve on teams with people in some
of the most complex domains. Since human teams often rely on
task-oriented dialogue to coordinate their actions, robot teammates
will no doubt need to rely similarly on this modality. However,
constraining factors such as workload, time pressure, and reduced
situational awareness are prevalent in these domains, and affect human
speech through increased speech rate, disfluency, and speech overlap.
Though typically viewed as "messy" features of speech that are either
ignored or parsed out of most dialogue systems, evidence from
Psycholinguistics and Conversation Analysis suggests that these
features are important resources in the interaction that can support
coordination. Thus, in order for robots to serve as effective
teammates, they will need to be able to handle (and in some cases,
produce) these features of natural language. In this talk, I will give
an overview of my work to date addressing this challenge.
First, I will discuss some empirical work using the Cooperative Remote
Search Task (CReST) corpus to identify dialogue and interaction
patterns in a search-and-rescue scenario involving human teams. I will
then show some of the novel mechanisms we have implemented in a
cognitive robotic architecture to enable robots to handle some of the
features seen in the corpus - in particular, overlapping speech.
Finally, I will discuss remaining work, including additional modelling
as well as evaluation scenarios for testing our proposed mechanisms
and for further studying human-robot teaming.