Invited Speakers & Panels
There will be one invited talk and two panels.
Monday, August 23, 11:00 - 12:00 Invited Talk
"The time has come", the Walrus said, "to talk of many things"
Professor Erik Sandewall
Linköping University, Sweden
The name of the field suggests that cognitive robotics shall be concerned with the principled design of robotic systems with cognitive capabilities. Such systems should be able to represent knowledge about themselves and their environment, to gradually accumulate more knowledge from sensor input information and from their experience when performing actions in the world, and to use this knowledge for intelligent, purposeful behavior. The intelligence of the system may be displayed both through its own actions, and through its dialogue with human operators or with other robots.
In practice, the sensor systems of contemporary robotic systems constitute a bottleneck. The information that the robot can obtain from its sensors, including its vision system, is rarely sufficient as a basis for particularly interesting cognitive behavior. This restricts the rate of progress in the field.
In this talk I will propose that the field may also take interest in 'hybrid' robotic systems that obtain some of their information from databases and from the operators, since this can considerably increase the range of cognitive level information in the robotic system. Robotic systems of this kind arise, in particular, in network contexts containing large numbers of autonomous vehicles and sensor stations of different kinds, together with multiple human users.
The following are some of the new challenges in the design and deployment of cognitive robots in this broader sense, which will be addressed in the talk:
- Ability to represent a broader range of information. The
traditional focus on the representation of actions and change must be complemented with a study of representations for a wide variety of static phenomena, for example as presently studied in description logics.
- Ability to administrate large knowledge bases, even when
they contain inconsistencies. The use of 'microtheories'
e.g. in the CYC system is one possible approach for this end.
- Ability to communicate information between cognitive robots as well as between robot and human. This calls for the use of high-level message formats: should KQML be replaced by XML or by OWL? It also calls for the use of natural-language dialogue between robot and human operator.
- Ability to maintain cognitive robotic systems over long periods of time, so that there is an opportunity for non-trivial knowledge accumulation by learning or by being taught. The speaker's 'software individuals' concept is proposed as a way of achieving this goal.
There is of course already a lot of work in other parts of AI that has contributed to these problems, including work in semantic networks, description logics, and qualitative reasoning. However, the use of these techniques in cognitive robots imposes both simplifications and additional problems. In particular it is important that all facilities are well integrated with the sensoric and performative systems of the robot, including both those that are available todayand those that can be expected in the future.
Monday August 23, 18:00 - 19:00 Panel Discussion
Competitive and Evaluative Frameworks for Cognitive Robotics Research
Moderator: Micheal Thielscher
Panel: Patrick Doherty, <?>
Tuesday, August 24, 12:30 - 13:30 Panel Discussion
On the Future of Cognitive Robotics: where are we now,where are we heading?
Moderator: Patrick Doherty
Panel: John-Jules Meyer, Bernhard Nebel, Erik Sandewall, Alessandro Saffiotti