During my sabbatical stay March to May 2014 at the
University La Sapienza in Rome,
Department of Computer, Control, and Management Engineering Antonio Ruberti
(DIAG)
I am teaching a number of hours of the AI Seminar,
Friday 12-15 hrs.
Except for the first slot (Mar 7), the pattern is:
I will give a presentation, largely based on own research,
in the first part until the lunch break;
and after the break, we will discuss a paper from the literature
whose contents put the first part into a broader and/or different perspective.
All students are expected to have read the paper.
The discussion will be guided by a pair of students who have prepared the paper
particularly well.
Seminar Topic
The broad topic of my part of the seminar is Plan-Based Robot Control.
The background to this is the desire to specify, inspect, and communicate
robot tasks on a high semantic level;
to have the robots design their own courses of action for executing such a given task;
and to execute it in a goal-directed way, responding smartly to their actual environment
(which may differ from the environment model used at planning time).
Plan-based robot control obviously includes the problems of representing and perceiving
the environment on a high semantic level.
Schedule
March 7th
- Introduction and Research Background.
- Let me start by introducing ... -- well, not so much myself,
but my background in terms of the research on which the rest of my seminar part is based.
Before the lunch break, the focus will be on the academic, theory-driven research;
after the break it will be on applications.
This application-directed research is mostly done in the Osnabrück branch
of the Robotics Innovation Center
of the German Research Center for Artificial Intelligence (DFKI).
[slides]
March 14th
- Topic: Semantic Mapping.
- Whereas generating geometric maps or occupancy maps for and by robots (SLAM)
is a well-understood topic,
we have only begun to understand how it works that a robot generate a map that includes
objects and structures in semantic categories.
The talk will introduce into the problem
and present some own prior work on semantic mapping based on point cloud data (3D laser scans).
[slides]
- Paper
- Nico Blodow, Lucian Cosmin Goron, Zoltan-Csaba Marton, Dejan Pangercic, Thomas Rühr, Moritz Tenorth, and Michael Beetz:
Autonomous Semantic Mapping for Robots Performing Everyday Manipulation Tasks in Kitchen Environments.
Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems (IROS), 2011.
[paper link]
March 21st
- Topic: Object Anchoring and Symbol Grounding.
- One of the many things that humans do with no conscious effort and that robots cannot
really do yet, is to connect symbols in a representation (language utterance, knowledge base)
with objects and concepts "out there", as perceived, e.g., by sensors.
The problem in full breadth is a philosophical one,
often addressed under the headline symbol grounding.
This will be covered by the paper to read.
Engineers should be aware of the symbol grounding problem in full generality;
however, for a robot, it makes sense to start with a simpler version, called object anchoring.
This will be presented, along with solution ideas, in the technical talk in the beginning.
[slides]
- Paper
- Stevan Harnad: The Symbol Grounding Problem. Physica D 42: 335-346, 1990.
[paper link]
March 28th
No seminar hour by me!
April 4th
- Topic: Approaches to Plan-Based Robot Control.
- Here we go right into the central point of the seminar.
Contrary to intuition, the problematic part on plan-based robot control is not so much to generate a plan,
but to execute and monitor it and to integrate the plan-based part of the control
into "the rest" of the control that needs to be running, too.
The technical talk will present an approach to using an affordance-based architecture
to addressing the execution problem.
The paper gives a more general AI perspective on the issues in plan-based robot control.
[slides]
- Paper
- Felix Ingrand and Malik Ghallab:
Robotics and Artificial Intelligence: a Perspective on Deliberation Functions.
AI Communications 1/2013.
[paper link]
April 14th (!!!)
- Topic: "Service Robotics".
- If you read management-level texts about the market prospects of robotics research,
you will most frequently be told that this prospect is in building "service robots"
for doing the household, for elderly care, for whatever.
I think this is wrong.
The situation seems much rather to be analogous to Computer Science,
where the largest part of processors and software is not incorporated by PCs etc.
and software running on them,
but on embedded processors in everyday-objects, like cars and household appliances
that you would normally not associate with Computer Science.
So the perspective is incorporating robotics technology into everyday objects and environments,
rather than building stand-alone "service robots" to do mundane tasks.
The talk will explore that issue.
The paper is a technical one, giving an example for "embedded robotics technology"
that makes use of plan-based robot control technology.
[slides]
- Paper
- Federico Pecora, Marcello Cirillo, Francesca Dell'Osa, Jonas Ullberg, and Alessandro Saffiotti:
A Constraint-Based Approach for Proactive, Context-Aware Human Support.
J. Ambient Intelligence and Smart Environments, 2012.
[paper link]
| Last changed: 2014-05-14 |
