WMLSR 2015 - Workshop on Machine Learning for Social Robotics

Workshop on Machine Learning for Social Robotics
In conjunction with the International Conference on Robotics and Automation (ICRA)
May 26th 2015 (full day) in Seattle, Washington
Important Dates:
15 March: Workshop paper submissions due
15 April: Workshop paper acceptance notifications
26 May: Workshop at ICRA in Seattle
Submissions:
We solicit papers describing theoretical ideas, algorithmic advances, user studies, or new evaluation methodologies related to machine learning for social robotics. A non-exhaustive list of relevant topics is:
- Reinforcement learning,
- Transfer learning,
- Human and social behaviour understanding,
- Active learning with humans-in-the-loop,
- Social path planning and navigation,
- Social signal processing,
- Social scene detection,
- Modelling social interaction with partially observable Markov decision processes,
- Machine learning approaches to human-robot interaction,
- Language or dialogue learning for human-robot interaction,
- Interpreting social signals and reinforcement signals, and
- Tools and (benchmark) datasets.
Papers should be submitted in PDF format, be no more than 6 pages long, and conform to the ICRA manuscript guidelines (http://icra2015.org/contribute/paper-submission). The reviewing process will be single-blind so anonymising the authors is not necessary. To submit, please email your papers to s.a.whiteson-AT-uva.nl by 15 March 2015.
Up-to-date information about the workshop is available at http://bit.ly/1HA1Cp4
Please contact Shimon Whiteson (s.a.whiteson-AT-uva.nl) with any questions.
Workshop Objectives:
Social robots operate in a multi-modal world and are faced with increasingly complex tasks. These tasks have typically been addressed using hand-crafted knowledge and action rules. While initial progress has been made using this manual approach, it has its limitations with respect to generalisation, domain adaptation and efficiency. Fortunately, the vast amount of available data in combination with the increasing sophistication of machine learning techniques opens up possibilities for automatic learning in social robotics.
In this workshop, we consider the unique challenges and opportunities that arise in developing machine learning techniques for social robotics. In particular, we consider two different scenarios in which humans and learning can interact in robotics. In the first, ``learning in the presence of humans", a robot must master a skill in an environment that contains humans. Typically, the task involves observing or interacting with those humans, including human-robot interaction and social navigation. In the second ``learning with humans in the loop", the task itself may not inherently involve humans but they are nonetheless involved in a teaching capacity during the learning process, e.g., via imitation learning, co-learning, or inverse reinforcement learning.
Machine learning challenges for social robotics can also be classified based on their focus on either perception or action. In perception-based tasks, the goal is to learn to understand something about the humans in the robot's environment, e.g., people detection and tracking, emotion recognition, human and social behaviour analysis, and social signal processing. In action-based tasks, the goal is to learn appropriate robot behaviour, e.g., socially normative navigation and body language.
Learning that involves humans also poses practical challenges for evaluation. Much work remains to be done to establish metrics and facilitate rigorous comparisons.
Confirmed Invited Speakers:
Sonia Chernova: “Advancing interactive robot learning to real world domains and real world users''.
Dizan Vasquez: “Planning based behavior modeling and learning for dynamic environments''.
Andrea Thomaz: Title to be determined.
Organizers:
Shimon Whiteson, University of Amsterdam
Ronald Poppe, Utrecht University
Vanessa Evers, University of Twente
Luis Merino, Pablo de Olavide University,
Stephen van Rump, Giraff Technologies