FJA@RSS2018 > Program

Friday, June 29th 

9:00-9:30 Cognitive Processes for Theory of Mind

Laura Hiatt, Research Scientist at the U.S. Naval Research Laboratory in Washington, DC.

When working with a human teammate, it is important to understand and accommodate not only task level behaviors of the teammate, but also process level behaviors and idiosyncracies. These low-level cognitive processes play a critical role in human cognition, such as the process of remembering a fact related to the task at hand. Importantly, these processes are also noisy, leading to outcomes such as forgotten facts or skipped steps in a task. Here, I argue that modeling these processes can lead to a theory of mind that accommodates inherent, low-level human variability, ultimately enabling teammates to perform more effectively.

9:30-10:00 Two data-driven approaches to joint social dynamics

Séverin Lemaignan, Senior Researcher, Bristol Robotics Lab, University of the West of England, UK. 

In order to achieve fluid human-robot joint actions in natural environments, we need our robots to both understand and participate to the dynamics of the social interactions. In this talk, I will discuss our on-going, data-driven, efforts to either recognise or generate social behaviours. I will first present SPARC, a new paradigm to teach a robot to behave autonomously in a high dimensional social environment. I will then discuss our first results in using neural nets to recognise complex human social interactions.

10:00-10:30 REFRESHMENT

10:30-11:00 What Is Theory of Mind, and Could It Help with Joint Action?

Tadeusz W. Zawidzki, Department Chair, Associate Professor of Philosophy at George Washington University, DC.

It is a common assumption that human social competence is made possible by a sophisticated theory of mind. However, there is surprisingly little consensus on what a theory of mind is supposed to be. Here, I argue that, on some conceptualizations of “theory of mind”, there are good reasons to think it cannot help in joint action or other examples of human-like social competence. When we try to re-conceptualize “theory of mind” in ways that might avoid these problems, it starts looking less like a “theory of mind”, and more like a set of capacities to track abstract behavioral patterns. I conclude with a hypothesis: supplementing such capacities with “reactive attitudes” that enforce norms of interactive behavior can yield central components of human-like social competence, like joint action.

 video of the talk

11:00-12:00 Hand-On Session

about benchmarks and how to define them... 

what are we trying to optimize?
find a game with several levels of abstraction?
which degree of embodiment? (collaborative -> social)
find several complementary examples 
need tasks with interaction...

12:00-13:30 LUNCH

13:30-14:30 Teaser + Poster Session

Addressing Challenges of Theory of Mind using Hybrid Conditional Planning

Momina Rizwan, Esra Erdem and Volkan Patoglu

Interactive Plan Explicability in Human-Robot Teaming (an extended version of this paper will be presented at RO-MAN'18)

Mehrdad Zaker Shahrak, Yu Zhang

Learning Human Navigational Intentions

Mahmoud Hamandi and Pooyan Fazli

Toward a Real-time Activity Segmentation Method for Human-Robot Teaming

Tariq Iqbal, Laurel D. Riek, Julie A. Shah

Gaze for Error Detection During Human-Robot Shared Manipulation

Reuben M. Aronson, Henny Admoni

Human Target Prediction in Cooperative Manipulation Tasks

Mahmoud Hamandi, Emre Hatay, and Pooyan Fazli

Implementing an implicit false belief task by cascading situation assessment with UNDERWORLDS

Yoan Sallami, Séverin Lemaignan, Aurélie Clodic, Rachid Alami

Modelling Epistemic States in Human-Robot Interaction: A Logical Approach

Emiliano Lorini, Fabian Romero, Rachid Alami, Aurélie Clodic

14:30-15:00 REFRESHMENT

15:00-15:30 Natural Human Behaviors for Theory of Mind

Henny Hadmoni, Assistant Professor, Robotics Institute, Carnegie Mellon University

15:30-16:00 Adaptation in Team Strategies for Implicit Communication

Ross A Knepper Assistant Professor, Department of Computer Science

In order to collaborate, teams must regulate consensus about common ground.  Teammates must agree on goals and appropriate actions to achieve them.  Communication mediates the maintenance of common ground.  However, communication is itself often ambiguous, occurring implicitly through actions occurring within the structure of the joint activity. In this talk, I describe a framework for understanding and generating implicit communication between a human and a robot.  This type of communication is properly thought of as encoding meaning in the manner of doing some functional action.  As a simple proxy for human-robot teaming, I study implicit communication in the collaborative card game Hanabi.  In this game, players can see every other player's cards but not their own cards.  To know what card to play, they must give each other hints through a highly constrained language. Although the language is restricted to communicating incomplete card information, it permits implicit communication that requests a player to take a specific action.  Most novice players eventually automatically discover this communication strategy, suggesting that it is an innate human ability.  However, we present evidence that adaptation to a consensus strategy is a crucial element of successful Hanabi gameplay.  We discuss the implications for human-robot teamwork and provide a teaser for a study in progress in the Hanabi domain.

 video of the Hanabi Game

16:00-17:00 Panel Session (moderator: Chien Ming Huang)

with the participation of 

    Henny Hadmoni, Assistant Professor, Robotics Institute, Carnegie Mellon University