Call for Participation
ROBOTICS 2005 Workshop on Modular Foundations for Control and Perception

in conjunction with Robotics: Science and Systems Conference 2005
Massachusetts Institute of Technology, Cambridge, MA, USA
June 11, 2005

Workshop Introduction and Discussion Topics

Advances in the field of robotics have begun to realize sophisticated robotic platforms with increasingly richer sensing requirements. In addition to increasing robot functionality, robotics researchers also strive towards increasing the accessibility of robot control to greater segments of society. The balancing of greater robot functionality versus greater robot accessibility can be difficult. In addressing these issues, greater emphasis has been placed towards creating and building upon basic modular components. Using neuroscience and biology as inspiration, such modules reduce of the complexity of the control and sensing interfaces while providing necessary functionality.

More specifically, we consider a module (or "primitive" or "behavior") to encodes a link between perception and motor control. This link furthers the computational process that enables a robot to achieve or maintain certain goals. A basis set of such primitives is sufficient, through various combination operators, for generating the entire movement/activity repertoire of a robot. As a consequence the development, learning and utilization of such primitives robot control has become an area of high interest in high DOF systems (such as humanoids), learning by imitation or demonstration and understanding of user activity or intent.

The workshop will explore modular foundations for robot control and perception and will address three basic areas:

1) development and learning of modular primitives
2) utilization of modular foundations for both control and perception
3) utilization of known primitives for symbol grounding in higher level methods

Motivation and Objectives:

The primary motivation for this workshop is to gain a broader picture of current research in learning and developing basic modular functions for robotic systems. In the course of pursuing high-level or task-oriented robot functions, efforts in robotics can overlook the principles and methods in developing low-level (or primitive-level) procedures. This workshop aims to bring together researchers working on developing, discovering, and utilizing low-level modules and exchange in discussion on unifying principles.

The objectives of this workshop are related to various topics in robotics and autonomous systems, including:

* learning by demonstration
* imitation learning
* data-driven analysis of robot teleoperation
* acquisition and refinement of task descriptions and primitive skills
* modular foundations linking both perception and action
* developing sensory-motor controllers
* predictive and dynamical models
* autonomous control of humanoid robots and high DOF systems
* activity recognition and user modeling
* reinforcement learning using modular states and actions
* symbol grounding for autonomous systems
* behavior cloning
* human-robot interaction
* brain-machine interfaces
* supervisory control
* behavior-based and hybrid control architectures
* mobile manipulation


9:00-9:25 Chad Jenkins Introduction / Performance-Derived Behavior Vocabularies  
9:25-9:50 Martin Giese Representation of Action Sequences by Linear Combination of Prototypes  
9:50-10:15 Rod Grupen Deriving Intentions using a Control Basis  
10:15-10:30 Break  
10:30-10:55 Auke Ijspeert Nonlinear Dynamical Systems for the Modular Control of Locomotion And Movement: Experiments With Pattern Generators  
10:55-11:20 Jan Peters / Stefan Schaal Learning Motor Primitives with Reinforcement Learning  
11:20-12:00 Poster Session  
12:00-1:30 Lunch  
1:30-1:55 Dana Ballard TBD  
1:55-2:20 Monica Nicolescu Learning from Demonstration. Generalization and Feedback Using Existing Primitive Skills  
2:20-2:45 Darrin Bentivegna Combining Local and Global Features to Support Generalization and Learning from Practice [PPT] [Articles]
2:45-3:15   Discussion / Break  
3:15-3:40 Hiromu Onda Visualization and Simulation of Sensory Events as a Representation of States for Primitive Skills [PPT] [Articles]
3:40-4:05 Emo Todorov
From Task Parameters to Motor Synergies: A Hierarchical Approximation to Optimal Feedback Control of Redundant Systems  
4:05-4:30 Alan Peters TBD  
4:30-5:00 Jenkins, Nicolescu Discussion/Workshop Conclusion  


This one-day workshop will consist of a several talks and a poster session on focused topics in modular perception and control. These presentations will be complimented panel sessions to facilitate open discussion of clusters of closely related talks.

Participation and Submission:

Submissions to this workshop can be extended abstracts (1-2 pages) or technical papers (not exceeding 6 pages in length). Submissions should be sent via email to Chad Jenkins (cjenkins[a-t]cs.brown.edu) and Monica Nicolescu (monica[a-t]cs.unr.edu) in PDF format. Submissions should identify concepts and positions about the state-of-the-art and provide clarity to the greater picture of modular perception and control. The workshop is open to submissions addressing related topics in the form of research, applications, position statements, or unifying surveys.

From these submissions, the workshop organizers will select presentations to appear at the workshop. All accepted presentations will require a poster presentation. Oral presentations will be selected from the set of accepted submissions. If the papers are of sufficient quantity and quality, we will seek to publish them as an edited book or journal special issue.

Important Dates:

March 26, 2005: Submission deadline for papers and abstracts
April 2, 2005: Notification of acceptance
TBD: Camera ready versions due
June 11, 2005: Workshop date

Workshop organizers:

Odest Chadwicke Jenkins

Monica Nicolescu

Roderic Grupen

Richard Alan Peters II