Chicago, IL, USA, July 13-17, 2008
_________________________________________________________________________________
Workshop on Mobility and Manipulation: July 14 (13:50 - 18:00)
This workshop explores advanced perception and cognition that significantly advances and/or speeds robot mobility and/or manipulation. Format: A distinguished panel will kick off the workshop by providing a broad overview of the area, emerging paradigms and exciting opportunities. Presentations will be 20-22 minutes not including time for questions. Those invited to exhibit at AAAI, will give short 10-minute briefs. These talks provide examples and personal thoughts of mobility and manipulation. The workshop concludes with a 60-min Discussion Forum to capture thoughts and generate research roadmaps.
Schedule
|
Time |
|
|
||
|
13:50 |
Monica Anderson |
Drexel University and Workshop Chair |
Opening Remarks |
|
|
14:00 |
DARPA |
IPTO Programs |
||
|
14:25 |
Microsoft Research |
Robotics Research |
||
|
14:50 |
University of Massachusetts, Amherst |
Results of 2005 NASA/NSF Workshop on Mobile Manipulation |
|
|
|
15:15 |
Ron Provine |
Boeing Phantom Works |
AI Robotics |
|
|
15:30 |
TBD |
|
|
|
|
16:00 |
Coffee Break |
|
|
|
|
16:25 |
Howie Choset |
CMU |
Snake Robots |
|
|
16:30 |
Stanford |
STAIR: Stanford AI Robot platform |
|
|
|
16:35 |
Southern Illinois University |
Formations and Legged Robots |
|
|
|
16:45 |
CMU |
Hexapods |
|
|
|
16:50 |
Kansas State University |
Semantic Vision |
|
|
|
16:55 |
Western Washington University |
Robot Vision |
|
|
|
17:00 |
All Participants |
Discussion Forum |
Research Roadmaps: Near-term and Grand Challenges |
|
|
18:00 |
Adjourn |
|
|
|
|
|
|
|
|
|
Exhibits will be showcased on July 15 and 16. This gives audiences opportunities to see and experience what was discussed in the Workshop.
Bios
Monica Anderson
_________________________________________________________________________________
Robert
Mandelbaum
Challenges in Mobile Manipulation
Abstract:
The success of tele-operated robots in defusing explosives in Iraq has
heightened interest in imbuing mobile robots with even greater dexterity and
with capabilities to manipulate their environments in a more autonomous way.
However, integrating and controlling multiple manipulators on mobile platforms
presents numerous simultaneous challenges in the areas of perception,
manipulator design, grasp planning, pose planning, and closed-loop control. In
this talk, I present several challenge problems which DARPA is interested in addressing.
Bio:
Dr. Mandelbaum
is a Program Manager at DARPA in robotics. He currently manages the BigDog and Learning Locomotion programs. Dr. Mandelbaum was previously the Director of the Automotive
Business Unit for Sarnoff Corporation. As Founder of Sarnoff's
Robot and Automotive Vision Group he developed and applied perception
technologies to Unmanned Ground Vehicles, and the commercial automotive
industry. Dr. Mandelbaum has a Ph.D. in Computer and
Information Science from U. Penn. He is a faculty member at Drexel University,
and has taught a graduate course on "Computer Vision for Robots". He
has published numerous technical papers in robotics, vision, and perception.
_________________________________________________________________________________
Stewart Tansley
Bio:
Stewart is
responsible for academic partnerships in Robotics research and Sensor Networks
research as part of External Research & Programs in Microsoft Research.
Before joining Microsoft in 2001, he spent 13 years in the telecommunications
industry in software research and development, focusing on technology transfer.
Stewart has a PhD in Artificial Intelligence applied to Engineering from Loughborough University, UK. He has published a variety of
papers on robotics for education, artificial intelligence and network
management, several patents, and has co-authored a book on software engineering
for artificial intelligence applications (so long ago that he should really
write a new one).
_________________________________________________________________________________
Autonomous Manipulation in Unstructured Environments
Abstract:
Autonomous manipulation in unstructured environments is a prerequisite for many
important applications, ranging from household robotics to urban search and
rescue efforts. We demonstrate autonomous manipulation of unknown objects
in the context of a specific task. Examples of such tasks could include
the opening of doors, hatches, and lids, or the use of tools such as pliers,
wire cutters, or scissors. The first demonstration is a method for
extracting the kinematic structure of unknown objects to enable manipulation.
The second demonstration is a skill for visual segmentation and tracking
of objects, again as a prerequisite for autonomous manipulation. The resulting
manipulation capabilities are highly robust. We therefore argue that our
approach to manipulation is well-suited for real-world mobile autonomous manipulation.
Bio:
Oliver Brock is an Assistant Professor of Computer Science at the University of Massachusetts Amherst. He received his Computer Science Diploma in 1993 from the Technical University of Berlin and his Masters and Ph.D. in Computer Science from Stanford University in 1994 and 2000, respectively. He was a co-founder and CTO of an Internet startup called AllAdvantage.com. He also held post-doc positions at Rice University and Stanford University. At the University of Massachusetts Amherst, Oliver is affiliated with the Robotics and Biology Laboratory and the Computational Biology Laboratory. His research focuses on Autonomous Mobile Manipulation and the application of robotic algorithms to problems in Structural Molecular Biology.
_________________________________________________________________________________
Ron Provine
_________________________________________________________________________________
Howie Choset
_________________________________________________________________________________
STAIR: Learning to Open New Doors
Abstract:
We consider robots operating in novel, uncertain home and office environments. In these environments, they should be able to open doors and elevators to navigate to new spaces. This, however, remains a challenging problem because a robot will likely encounter doors it has never seen before. We present a vision-based learning algorithm that finds door handles and elevator buttons, and also infers a strategy to open them. This enables our robot to navigate anywhere in a new building by opening doors and elevators, even ones it has not seen before. We also present our learning algorithms that enable our robots to grasp novel objects.
Bio:
Andrew Y. Ng received his B.Sc. from Carnegie Mellon University, his M.Sc. from the Massachusetts Institute of Technology, and his Ph.D. from the University of California, Berkeley. He is an Assistant Professor of Computer Science at Stanford University, and his research interests include machine learning, robotic perception and control, and broad competence AI. His group has won best paper/best student paper awards at ACL, CEAS and 3DRR. He is also a recipient of the Alfred P. Sloan Fellowship.
_________________________________________________________________________________
SkewlZone: A Brain and Sensor Pack for Advancing Research and Education in Legged Robot Mobility
Abstract:
Legged robots with sufficient on-board processing power for real-time sensing and control are not currently available on the market. SkewlZone is a generalized system that provides perception and cognitive ability to off-the-shelf legged robots such as the Kondo KHR-2HV. Specialized haptic foot sensors, hand sensors, and inner-ear balance (3-axis inertial measurement unit) are integrated with an on-board embedded XScale Linux board using a standard I2C interface. The software architecture allows real-time sensing and control on-board. Applications can be quickly developed using the open interface, which provides a layer of abstraction away from the details of the mechanical control.
Road Runner: The Smart Cart
Abstract:
The Mini Grand Challenge competition held at Penn State Abington is scaled to demonstrate autonomous vehicle technologies on low-cost platforms. Thus far entries have been too small to transport a person. Our single-person golf cart entry, Road Runner, allows a passenger to ride comfortably and interact via a laptop interface. We leverage Google Earth as a means to specify waypoints. As the robot traverses these waypoints, it provides information regarding known landmarks. The Road Runner provides a platform to test autonomous vehicle interfaces and gain insight into the ways that humans will interact with the automobile of the future.
A Distributed Control Algorithm for Robots in Grid Formations
Abstract:
Coordinating a group of robots to work in formation has been suggested for a number of tasks. Our approach to formation control treats each robot as a cell in a cellular automaton. A robot's behavior is governed by a set of rules for changing its state with respect to its neighbors. Using only local communication and sensor readings, robots calculate and correct for discrepancies between desired and actual relationships, producing movements that result in the emergent organization of the desired global structure. This work extends on our previous work by considering grid formations, such as square and hexagonal lattices.
Robotic Limb Calibration: Accelerometer Based Discovery of Kinematic Constants
Abstract:
Limbed robots carry inherent difficulties not found in wheeled or tracked robots. Robots with linkages typically require the use of careful measurements and complex kinematic equations for set-up and calibration. In our research we aim to automate this process through the use of feedback from a 3-axis accelerometer and algorithms that solve for the linkage lengths. The calibration involves the structured movement of the limb(s) and data collection from the highly sensitive accelerometer as well as calculation of the kinematic equations. Benefits of this research include simple set-up, constant feedback and interesting possibilities in self-discovery.
Bio:
Jerry B. Weinberg is a Professor in the Computer Science Department at Southern Illinois University Edwardsville. He teaches courses and conducts research in Artificial Intelligence, Robotics, and Human-Computer Interaction. In 1999, Dr. Weinberg formed the Robotics Project Group that has introduced robotics projects in various computer science and engineering courses, and has initiated various robotics outreach programs. He was the Principal Investigator on the NSF funded projects "An Undergraduate Robotics Course Emphasizing Integrated System Design and Multidisciplinary Team Work" and "The Effects of Robotics Projects on Girls' Perceptions of Achievement in Science, Technology, Engineering, and Mathematics".
_________________________________________________________________________________
David Touretzky
A New Hexapod Robot for Robotics Education
Abstract:
Since Sony discontinued the AIBO in 2006, there has been no comparable platform for robotics education. We are developing a hexapod robot that retains many of the AIBO's best features: a powerful onboard computer, built-in video camera, servos with position and force feedback, and wireless Ethernet for remote monitoring and teleoperation. Our platform surpasses the AIBO in two ways. With six legs instead of four, the robot's walk is statically stable. And the 6DOF arm allows it to grasp and manipulate objects. The robot is programmed using our Tekkotsu open source software framework. Anticipated cost is $2500.
Bio:
David S. Touretzky is Research Professor of Computer Science and co-director of the graduate training program of the Center for the Neural Basis of Cognition at Carnegie Mellon University. His current research efforts aim to change the way undergraduate CS majors are introduced to robotics, by providing high level primitives for perception, manipulation, and control in the Tekkotsu programming framework. Dr. Touretzky's other major research area is computational neuroscience, focusing particularly on spatial representations in the rodent hippocampus and related structures. He received his Ph.D. from Carnegie Mellon in 1984.
_________________________________________________________________________________
Semantic Vision
Abstract:
The semantic vision challenge in 2007 inspired our robotics team to include more intelligence in our robotic system. Instead of building intelligence into the controller software, our team has been investigating ways to use information from the internet to build internal models that allow the robot to make better sense of the assigned tasks and improve object recognition.
Bio:
David Gustafson is a Professor at Kansas State University in the Computing and Information Sciences. He earned his PhD from the University of Wisconsin - Madison. His research interests include robotics, vision, biometrics, software engineering, software metrics and software testing. His teams have participated in many AAAI Robotics competitions since 1995.
_________________________________________________________________________________
A Seeing-Eye Guide Robot Prototype Using NXT
Abstract:
Service animals undergo rigorous, lengthy training to fill many difficult roles for the benefit of their owners. A subset of these roles that depend on the ability to learn and intelligently respond to a variety of external stimuli can also be filled by robots with less training time and maintenance cost. This paper explores such an approach with a "Seeing-eye Guide Robot" that is trained by feedback from both humans and the environments using parallel learning models. These models will allow the robot to selectively obey human commands depending on its understanding of the safety. The prototype of seeing-eye guide robot is made of NXT Lego Mindstorms robot kit. We expect the learning results satisfy our goals before implementing a more sophisticated model.
Bio:
Dr. Jianna Zhang is an associate professor at Computer Science Department, Western Washington University. Her research interests falls in machine learning, robotics, natural language processing, and Web design. Dr. Zhang is the president of Bellingham AI and Robotics Society, the supervisor of Western Student Robot Club, and the creator and project supervisor of ML and Robotics Lab at Computer Science Department of Western Washington University.