List of Invited Speakers

Oliver Amft

Eindhoven University of Technology

Interpreting human behavioural patterns using ubiquitous systems

Sensors-equipped ubiquitous systems, such as smartphones are an excellent example of how ubiquitous technology becomes omnipresent in daily life of many people. Still, context-aware and user behaviourrelated functionalities of ubiquitous systems are pending to reach the consumer market. In this talk, I will shed some light on established machine-based context and activity interpretation concepts. Recent results aiming at improving robustness, generality, and intelligibility will be discussed. In particular, I will focus on how algorithms can cope with complex human behaviour in real life. This this end, key fundamental challenges in activity and context recognition must be addressed, pertained to diversity of activities, concept drift, and missing reference standards. Showcases of recent works in multimodal activity recognition are presented related to personal assistance in chronic diseases, wellness guidance, and smart building automation.

Chatschik Bisdikian

IBM Research, T. J. Watson Research Center

On the Quality of Information for Sensor-Enabled Systems and Solutions

In the highly heterogeneous, information rich environments envisaged in emerging smart environments, disseminating sensor-originated information with desired quality characteristics is key to the effective execution of high-tempo dynamic tasks. A characterization of the quality of information (QoI) is useful in many contexts and can be invaluable in understanding our sensed surroundings and taking effective actions in response. Understandably the manner of interpreting and representing the QoI is highly application-dependent hampering streamlining QoI-aware applications. An application-agnostic QoI specification can provide consistency in the representation of information and its quality, and enable QoI-aware determinations across many different applications. In this talk, we present a few QoI motivating sensor-enabled applications, then build a layered definition of QoI, discuss its implications, and conclude with some of a recent research work in the area.

Tully Foote

Willow Garage

Tully Foote is a Systems Engineer at Willow Garage.  He is the ROS Platform Manager.  Recently he has been focusing on extending the support of ROS and the infrastructure to more platforms and architectures and user groups. As part of this work he was a co-creator of the TurtleBot project.  Previous to working at Willow Garage, he worked on autonomous cars in all three of the DARPA Grand Challenges, first at Caltech and then at University of Pennsylvania for the Urban Challenge.

Randy Freeman

Department of Electrical Engineering and Computer Science, Northwestern University

I) Decentralized filters for the robust averaging of time-varying signals in mobile sensor networks

Several algorithms for the control of mobile sensor networks rely on the ability to compute global averages of local signals in a decentralized, scalable, and robust manner. However, many of the simplest schemes for distributed averaging lack a key robustness property and can therefore exhibit unwanted steady-state errors. In this talk, we present necessary and sufficient conditions for members of a broad class of decentralized filters to succeed in robustly computing exact global averages of time-varying signals.

II) Applications of decentralized averaging in mobile sensor networks

This talk outlines some applications of distributed averaging in mobile sensor networks, including environmental monitoring via decentralized Kalman filtering and decentralized estimation and control of algebraic connectivity.

Lynne Parker

The University of Tennessee

Dynamic Task Allocation in Heterogeneous Robot Teams

In a heterogeneous team of robots, the morphological and/or behavioral capabilities of the team members can vary significantly from robot to robot. These differences lead to very different abilities of robots to perform the tasks within an application. A key challenge in these problems is to determine the proper allocation of tasks to robots, to attempt to optimize some criteria of performance. Unfortunately, this problem is similar to the set-covering problem, which is an NP-hard problem. Thus, researchers have developed a variety of alternative approximation approaches to task allocation to address this problem. In this tutorial, we will define the general multi-robot task allocation problem (MRTA), and discuss a variety of alternative techniques for addressing this problem. Approaches to be discussed include behavior-based approaches as well as market-based approaches. The intent of this tutorial is to outline the general strategies that have been proposed, along with comparisons of alternatives.

Jan Peters

TU Darmstadt

Towards Learning to Compete and Cooperate

As Jan Peters' research lies at the intersection between two fields, i.e., machine learning and robotics, he has always keen to bring members of both fields together.


Kay Römer

University of Lübeck

Fault Detection in Networked Embedded Systems Deployments

Networked embedded sensing systems are typically exposed to a hostile and unpredictable environment, often leading to failures despite extensive pre-deployment validation in testbeds. Once deployed, these systems are hard to debug due to limited acccess to the deployment site and due to constrained system resources. This talk will survey typical failures encountered, look at fundamental techniques to detect them, and discuss concrete fault detection systems that have been developed.