From mahadeva@guardian.csee.usf.edu Tue Mar 29 22:28:51 EST 1994 Article: 21269 of comp.ai Xref: glinda.oz.cs.cmu.edu comp.ai:21269 Path: honeydew.srv.cs.cmu.edu!das-news.harvard.edu!noc.near.net!MathWorks.Com!europa.eng.gtefsd.com!gatech!udel!darwin.sura.net!mother.usf.edu!guardian!mahadeva From: mahadeva@guardian.csee.usf.edu (Sridhar Mahadevan) Newsgroups: comp.ai Subject: MLC-COLT '94 Workshop on Robot Learning Date: 22 Mar 1994 19:53:36 GMT Organization: University of South Florida Lines: 141 Sender: mahadeva@guardian (Sridhar Mahadevan) Distribution: world Message-ID: <2mnic0$fj3@mother.usf.edu> NNTP-Posting-Host: guardian.csee.usf.edu Keywords: robotics, machine learning, AI --------------------------------------- MLC-COLT '94 WORKSHOP ON ROBOT LEARNING July 10th 1994 Rutgers Univ., New Brunswick, NJ ---------------------------------------- DESCRIPTION OF WORKSHOP: ------------------------ Building a learning robot has been acknowledged as one of the grand challenges facing artificial intelligence. Robotics is an extremely challenging domain for machine learning: sensors generate huge noisy state spaces, actions are often continuous, and training time is very limited. The area of robot learning has witnessed a surge of activity recently, fueled by exciting new work in the areas of reinforcement learning, behavior-based robotics, genetic algorithms, neural networks, and artificial life. The goal of this workshop is to discuss/communicate research ideas from these and other relevant fields on building learning robots. This workshop will focus on tasks and learning methods appropriate for real robots. Many techniques look promising on simulations, but fail to work on real robots because they take too long, or require information that cannot be computed reliably from sensors. This workshop will also discuss a common format for making training data available across the Internet to enable researchers without access to real hardware to contribute their ideas and expertise to robot learning. We encourage researchers from several different research areas, including robotics, machine learning, planning, vision, and neural nets to attend this workshop. This diversity will facilitate interaction among disparate research groups who face similar problems. Researchers with knowledge of real robots are especially encouraged to attend to share their experience. TOPICS: ------- The workshop will include (but is not limited to) the following topics: 1. ADAPTIVE CONTROL ARCHITECTURES: This includes work on learning new behaviors, learning to coordinate existing behaviors, and learning in hybrid declarative-reactive architectures. 2. PERCEPTUAL LEARNING: This includes building categories from sensor data, such as learning to recognize visual objects, or clustering sonar data. 3. LEARNING SPATIAL MAPS: This includes work in the area of robot navigation. 4. LEARNING ARM CONTROL AND MANIPULATION: This includes learning to control robot arms, and learning to pick up objects. 5. LEARNING WITH DOMAIN-SPECIFIC BIAS: Many learning researchers have come to the conclusion that tabula rasa learning is not effective for robots. If that is so, we must add domain-specific bias in some form. What should it be? Examples include teaching, pre-specified behavioral structure, and pre-specified low-level behaviors. 6. COMPARATIVE STUDIES: Lastly, the workshop will also encourage comparative studies among different learning methods, particularly those done using real robots. FORMAT OF THE WORKSHOP: ----------------------- The workshop will begin with an overview of the current state of the field, followed by short presentations of current research. There may be one or two invited talks if time permits. There will be a video session featuring demos of learning robots. Finally, there will be a panel discussion to summarize what we've learned and discuss issues for future research. We plan on also getting together for a workshop dinner after the workshop. CALL FOR PAPERS: ---------------- Researchers interested in presenting their work are encouraged to submit papers to the workshop. Please follow the same formatting guidelines used in the regular ML conference (i.e. 12 pages in Latex 12 pt article style excluding title page and bibliography, but including all tables and figures). Four copies of each paper should be mailed to the Chair (see address below). The deadline for receiving papers is May 1. Authors of accepted papers will be notified by May 21st, and camera-ready papers should be submitted by June 15th. We will try to distribute the workshop proceedings before the workshop. CALL FOR VIDEOS: ---------------- We also encourage researchers to submit short videos featuring their robots. We will accept all videos if they are reasonably short (no more than 10 minutes). Depending on the number of submitted videos, and the time constraints for the overall length of the video session, further editing may be required. ATTENDANCE: ----------- Attendance in this workshop will be by invitation only. All researchers interested in attending this workshop should submit a one page abstract outlining their current and previous work in robot learning. In order to maximize the degree of interaction among the participants, we will give priority to authors of accepted papers, invited speakers and panelists, and researchers who have had experience in working with real robots. PROGRAM COMMITTEE: ------------------ Leslie Kaelbling Sridhar Mahadevan (Chair) Box 1910 Dept of Computer Science and Engineering Dept. of Computer Science University of South Florida Brown University Tampa, FL 33647 Providence, RI 02192 mahadeva@csee.usf.edu lpk@cs.brown.edu Maja Mataric Tom Mitchell MIT AI Lab School of Computer Science Cambridge, MA 02139 Carnegie-Mellon University maja@ai.mit.edu Pittsburgh, PA 15213 mitchell@cs.cmu.edu -- ------------------- Sridhar Mahadevan Department of Computer Science and Engineering University of South Florida 4202 East Fowler Avenue, ENG 118 Tampa, Florida 33620-5399 mahadeva@csee.usf.edu Phone: (813)974-3260 Fax: (813)974-5456