From kube@cs.ualberta.ca Tue Jun 7 17:27:01 EDT 1994 Article: 11104 of comp.robotics Xref: glinda.oz.cs.cmu.edu comp.robotics:11104 Path: honeydew.srv.cs.cmu.edu!bb3.andrew.cmu.edu!news.sei.cmu.edu!cis.ohio-state.edu!math.ohio-state.edu!howland.reston.ans.net!gatech!newsxfer.itd.umich.edu!nntp.cs.ubc.ca!alberta!kube From: kube@cs.ualberta.ca (Ron Kube) Newsgroups: comp.robotics Subject: June's Grad. Students Who's Who in Robotics Date: 6 Jun 1994 23:35:36 GMT Organization: Computing Science, U of Alberta, Edmonton, Canada Lines: 1093 Message-ID: <2t0bs8$154@scapa.cs.ualberta.ca> NNTP-Posting-Host: kitscoty.cs.ualberta.ca Keywords: monthly posting June 6, 1994 >>>>>>>>>>>>>> GRAD STUDENTS WHO'S WHO IN ROBOTICS <<<<<<<<<<<<<< ================================================================= Have you ever wondered what grad students are doing in robotics? A trip to your local research library allows you to see Who's Who in robotics at the post-doc level, i.e. Professor SoNso, and SuchNsuch, but what about the graduate students working on their MSc. or PhD? Here is a summary of the received entries to date. If you would like to appear in the Grad Students Who's Who in Robotics send a note to kube@cs.ualberta.ca using the 5 point format. If you have WWW home-page then include its URL after your name. 1. Name: Arvin Agah email: arvin@robotics.usc.edu 2. Supervisor: George A. Bekey email: bekey@robotics.usc.edu 3. Institution: University of Southern California Institute for Robotics & Intelligent Systems 4. Research Area: Multi-Robot Systems & Biologically-Inspired Robotics 5. Summary: Investigating the issues of group behavior in robot colonies, both in simulation and hardware 1. Name: Peter O Aberg email: tm90pa@hh.se Anders Borghed email: tm90ab@hh.se 2. Supervisor: Per-Arne Wiberg email: pelle@hh.se 3. Institution: Centre for Computer Architecture, Halmstad University, Sweden 4. Research Area:Future Robotic Control 5. Summary: The direct and inverse kinematic problems are fundamental issues in robot modelling and control. The direct kinematics is a mapping which allows the position of the robot in the cartesian space to be related with the position in the joint space, i.e. given the position in the joint coordinates, we can univocally derive, by means of the direct kinematics, the corresponding position in the cartesian coordinates. The kinematic problem is different depending on the structure of the robot. For open-chain robotic structures the inverse kinematic poses the most difficulties and for closed-chain robotic structures it's the opposite. There are many different techniques for solving the kinematic problems. Most methods used today are based upon traditional mathematics, such as Newton-Raphsons method. None of these methods are in any way perfect for the problem. We have analyzed several different mechanical structures and control strategies that are in use today. In our work we also present methods to improve robotic motion performance by using new control methods and new mechanical structures. 1. Name: Karl R Altenburg email: altenbur@plains.nodak.edu 2. Supervisor: Mark Pavicic email: pavicic@plains.nodak.edu 3. Institution: North Dakota State University, Fargo, ND, USA 4. Research Area: Multiple Mobile Robots 5. Summary: Investigating the efficiency gains provided by communication and memory during multirobot search and retrieval type tasks. Currently tests are being conducted on a set of six small mobile robots, and in simulation. The work also investigates reactive control for individual robots and emergent control for the system. 1. Name : Venkateswara Rao Ayyadevara (email : avrao@vax2.concordia.ca) 2. Supervisors : Dr.R.M.H. Cheng (email : richard@vax2.concordia.ca) Dr.Ramesh Rajagopalan (email : ramesh@vax2.concordia.ca) 3. Institution : Centre for Industrial Control, Dept. of Mechanical Engineering Concordia University Montreal, Canada 4. Research Topic : Development of an Automated Robotic Deburring Workcell for Impeller Blades 5. Summary : Impeller is a component used in aircraft engines. Owing to their geometrical complexity and the rigorous standards specified by air safety regulations, impellers are extremely expensive to produce. After several thousand hours in use, the blades of an impeller are warped and the edges are corroded. If some of these impeller blades can be refurbished after being used for several thousand hours, considerable amount of money could be saved. I am working as part of a team which is developing an automated workcell using Yamaha Zeta Deburring robot to probe the surface of the impeller blade, reconstruct the surface, determine the desired edge profile and then use the robot to machine the workpiece to obtain that edge profile. My task is to develop set up for probing the surface of the impeller blades and to interface the controller of the robot with a PC-transputer network which is responsible for directing the probe, surface reconstruction and control of the robot. 1. Name: Tucker Balch email: tucker@cc.gatech.edu 2. Supervisor Ronald Arkin email: arkin@cc.gatech.edu 3. Institution Georgia Tech 4. Research Area: Communication in Autonomous Robot Societies 5. Summary: Multiple cooperating robots are able to complete many tasks more quickly and reliably than one robot alone. Communication between the robots can multiply their capabilities and effectiveness, but to what extent? In our research, the importance of communication in reactive robotic societies is investigated through experiments on both simulated and real robots. So far, our research has shown that for some tasks communication can significantly improve performance, but for others inter-agent communication is apparently unnecessary. In cases where communication helps, the lowest level of communication is almost as effective as the more complex type. Research is being extended to more complex scouting tasks for the Army. 1. Name: Johan G Benade email: jgb@ing1.rau.ac.za 2. Supervisor: Andre L Nel email: aln@ing1.rau.ac.za 3. Institution: Rand Afrikaans University, Johannesburg, RSA. 4. Research Area: Autonomous Robotics 5. Summary: Research is aimed at producing an improved biologically based controller for use in hexapod locomotion. The leg controller must be able to cope with uneven terrain - gaps in surfaces - inclines and surface tension variability. At the end of the project a functioning hardware realisation must be produced. 1. Name: Todd M. Bezenek email: bezenek@plains.nodak.edu 2. Super: Mark Pavicic email: pavicic@plains.nodak.edu 3. Istit: North Dakota State University, Fargo, ND 4. Area: Communications for multiple, autonomous robots. 5. Summary: Several groups are working with multiple robots to collectively solve a single problem. Those addressing the problem of communication between robots are assuming that there exists an unbreakable data path between each pair of robots, or between each robot and a central station. In many real applications where multiple robots may be used, communication between each pair of robots may not be continuous. As the robots move, the network representing pairs of robots that are able to successfully communicate changes. My goal is to develop a protocol which will allow robots on this network to communicate effectively. I have built two robots which communicate at 1200 baud over a simplex 49Mhz data channel. A third, which will act as a slave attached to a PC, is currently being constructed. 1. Name: William Chesters (http://www.dai.ed.ac.uk/students/williamc/williamc.html) 2. Supervisors: Gillian Hayes, John Hallam 3. Institution: Department of AI, University of Edinburgh 4. Research Area: Robot learning with neural networks 5. Summary: Neural nets look like an interesting approach to `bottom-up' AI, but if you try to apply them to non-trivial robot tasks, you come up against some serious problems: - a net tends to forget old knowledge as new experiences come in - it tends to get stuck in `local minima' - it can only work at one timescale: it can't support hierarchies of behaviours I'm interested in getting round these problems by using a community of competing nets. 1. Name: Howie Choset email: choset@robby.caltech.edu 2. Supervisor: Joel W. Burdick email: jwb@robby.caltech.edu 3. Institution: California Institute of Technology 4. Research Area: Sensor Based Planning for Mobile and Hyper-redundant Robots. 5. Summary: ``Sensor Based Planning'' incorporates sensor information, reflecting the current state of the environment, into a robot's planning process, as opposed to classical planning, which assumes full knowledge of the world's geometry prior to planning. Sensor based planning is important because: (1) the robot often has no a priori knowledge of the world; (2) the robot may have only a coarse knowledge of the world because of limited memory; (3) the world model is bound to contain inaccuracies which can be overcome with sensor based planning strategies; and (4) the world is subject to unexpected occurrences or rapidly changing situations. Currently, we are working on some initial steps towards path planning in a static environment where there is no a priori knowledge. We are develop- ing an incremental method to construct a Generalized Voronoi Graph (GVG), which is a 1-dimensional retract of a bounded space. The GVG is the same thing as a Generalized Voronoi Diagram in two dimensions. Like many other path planning schemes, the distance function is an integral part of the GVG. This function is nonsmooth; it is shown that the non-smoothness occurs at points which are ``critical'' to many path planning schemes. We have done some nonsmooth analysis on the distance function which has lead to the incorporation of simple and realistic sensor models. 1. Name: Chris Connolly email: connolly@cs.umass.edu 2. Supervisor: Rod Grupen 3. Institution: Laboratory for Perceptual Robotics, University of Massachusetts, Amherst, MA (USA) 4. Research Area: Motor and task planning using harmonic functions 5. Summary: Harmonic functions are solutions to Laplace's equation, and can be rapidly computed using resistive networks. They exhibit no local minima, and can be used to generate smooth goal-reaching trajectories [1]. We're using them for coarse reaching (on P-50 hand/arm systems) and mobile robot trajectory planning (on an unmanned ground vehicle). The resistive network formulation also turns out to be useful for modeling certain nuclei of the basal ganglia [2,3], and provide a theory for aspects of motor and cognitive planning in the mammalian central nervous system. [1] Connolly CI, Grupen RA, (1993) "The Applications of Harmonic Functions to Robotics", Journal of Robotic Systems, 10(7):931-946. [2] Connolly CI, Burns JB, (1993) "A Model for the Functioning of the Striatum", Biological Cybernetics, 68(6):535-544. [3] Connolly CI, Burns JB, (1993) "A New Striatal Model and its Relationship to Basal Ganglia Diseases", Neuroscience Research, 16:271-274. 1. Name: Joe Cronin J.Cronin@UNSW.edu.au 2. Supervisor: Richard Frost Richard Wilgoss 3. Institution: University of New South Wales, Sydney, Australia. 4. Research Area: Biped Robot. 5. Summary. I'll make it brief. It will have two legs. It will be anthropomorphic. It will walk. The ultimate goal of this project is to design, model and build a biped robot platform, capable of dynamic motion. It will be hydraulically driven, use an HC11 on every joint and stand about four feet high. There are two main areas of research; the first is to develop and control an ankle and foot with all degrees of freedom of the human ankle and foot, the second is to use distributed HC11's to solve the inverse kinematics at the joint. The project is at the stage where construction will begin before the end of 1993. As the school is in some financial difficulty, I have had to raise all funds for this project privately. The school does not have an active mobile robot group, I would be interested in anyone who would be interested in me. 1. Name: Bruce Digney digney@dvinci.usask.ca 2. Supervisor: M. M. Gupta 3. Institution: Dept. of Mech. Eng., University of Saskatchewan (Canada) 4. Research Area: Distributed Adaptive Control Systems 5. Summary: In my research I propose that by incorporated learning and adaption into a behavior based control system, the skills and behaviors which are impossible or impractical to be predetermined and embedded can be learned by the robot during operation. The result is a distributed adaptive control system (DACS), which can be thought of as the robot's artificial adaptive nervous system. This DACS autonomously learns the sensory-response couplings between the highest behavioral level, where the desired tasks are specified, and the lowest level actuators, which ultimately perform those tasks. A DACS has been developed for a simulated quadruped mobile robot and extensions to a physical robot are planned. 1. Name: Hans Dulimarta email: dulimart@cps.msu.edu 2. Supervisor: Anil K. Jain email: jain@cps.msu.edu 3. Institution: PRIP Laboratory Department of Computer Science Michigan State University East Lansing, Michigan 48824 4. Research Area: Distributed Robotics 5. Summary: I address the problem of task decomposition in a mobile robot navigation system. The underlying supposition of my approach is that in a typical robot navigation system, there are a number of modules running concurrently and each module is assigned a specific subtask. In order to accomplish the common goal of the navigation task, these modules share the resources and common data in the system. In such a system, resource access control and information sharing among the modules must be properly managed. In this work, I propose a decomposition of a robot navigation system into a number of {\it client} and {\it server} modules. Resource access control, resource sharing, information sharing, and process synchronization in the entire robot navigation system are delegated to the server modules. The clients send appropriate requests in order to avail of these facilities. There are two types of server modules defined in the system: {\it data server} and {\it hardware server}. The core of the system consists of one data server and several hardware servers. The data server acts as a common information exchange medium for all the clients, while the hardware servers provide access interface to the hardware or peripherals on the robot. By decoupling the hardware access routines from the hardware servers, the modules in the navigation system can be made independent of hardware platform being used. 1. Name: Sean P. Engelson email: engelson@cs.yale.edu 2. Supervisor: Drew V. McDermott email: mcdermott@cs.yale.edu 3. Institution: Yale University 4. Research Area: Map Learning 5. Summary: My work explores a `passive' mapping paradigm, in which the map-learning system has no direct control over the agent's actions. The main problem in map-learning is the fact that the agent's location is never perfectly known. Errors in localization lead inevitably to mapping errors. Passive mapping exacerbates this problem, since the mapper cannot perform experiments to verify the robot's location. My approach allows mapping errors to occur, and deals with them in two ways. First, is the use of a graph-based representation scheme which incorporates both connectivity and positional information to locally bound mapping error. Second, errors are diagnosed and repaired as information becomes available. The diagnosis and repair strategies are based on a taxonomy of possible mapping errors. 1. Name: Ted C. Feltmeyer email-tedhead@csd4.csd.uwm.edu email-felt1512@watt.cae.uwm.edu 2. Supervisor: Robert Borchelt email-borchelt@convex.csd.uwm.edu (Dr. Bob) 3. Institution: University of Wisconsin Milwaukee, Milwaukee, WI 4. Research Area: Automated diagnosis and error recovery in robotic workcells using artificial intelligence methods 5. Summary: I have worked since August 93 setting up a functional robotic workcell with one robot acting as a slave to the other. At this stage, the cell will perhaps perform some sort of electronic assembly operations. This physical system will become a testbed for different AI control techniques. My thesis will come out of the physical system setup and initial control using a hybrid expert system. My dissertation will expand further the different control techniques, plus integrate vision technology. The physical system is based on two Adeptone robots, AB PLC2/30, and (with luck) a new Pentium based PC. 1. Name: Johan Forsberg email: jf@sm.luth.se 2. Supervisor: Ake Wernersson 3. Institution: Robotics & Automation Lulea University of Technology S-971 87 LULEA, SWEDEN 4. Research Area: Autonomous Robots 5. Summary: I'm currently working with statistical map representation where the robot builds the map autonomously from measurements taken by a scanning range measuring laser. This statistical map is mainly intended for localization, while some other approach might be better for path planning. I think that to create an autonomous robot, we should not try to mimick the way humans, or animals, work. A better approach is to look at what computers are good at, and how this can be used to make a better robot. One example: when entering a room, we can know which room we are in either by recognizing the way the room looks, or by knowing exactly what route we took to get there (in meters and millimeters). The first method is the main method for humans, but extremely difficult for a computer. The second is difficult for a human, but easy for a robot. 1. Name: Bill Gribble email: grib@mamba.asg.arlut.utexas.edu 2. Supervisor: Ben Kuipers email: kuipers@cs.utexas.edu 3. Institution: University of Texas AI Lab 4. Research area: Distributed architectures for visually guided mobile robots 5. Summary: A robotic platform capable of manipulating increasingly large amounts of sensory data while still meeting real-time performance constraints must diverge from a monolithic model of computation. With a moderate number of compute nodes, each specialized for a task, the problem of sensor data partitioning can be more easily attacked. We are constructing a robot consisting of a number of nodes, including general-purpose microprocessors, small microcontrollers, and DSP engines, each with its own purpose, all linked by a 32Mbps custom local network. The software architecture is based on a packet protocol similar to that used in dataflow parallel machines. 1. Name: Bridget Hallam 2. Supervisor: Gillian Hayes 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Controlling Robots using Biological Theories 5. Summary: Studying animal behavioural control can give insights into autonomous behaviour that may prove useful for those wishing to build autonomous robots. Implementing Halperin's neuro-connector model of learning and motivation on a mobile robot has shown that it can be used to control robots, and that it is reasonably complete. Implementation in simulation will discover the sensitivity of the various features to variations in parameters and the exact equations used, and so improve the model as a robot controller. It may also improve the model for ethologists. 1. Name: Roger B. Hertz (hertz@ecf.toronto.edu) 2. Supervisor: Peter C. Hughes (hughesp@ecf.toronto.edu) 3. Institution: University of Toronto Institute for Aerospace Studies 4. Research Area: Articulated-Truss Manipulators 5. Summary: We are investigating the use of articulated truss mechanisms for both space and terrestrial robotics applications. We have constructed a prototype manipulator based on this concept that is capable of 3-DOF spatial motion. My research is centered on applying the technology to a 6-DOF industrial version of the manipulator. Current work is involved with manipulator design, development of kinematics algorithms, workspace analysis, and customization of an industrial robot contoller. 1. Name: Tomas Hogstrom email: tomas@idefix.ikp.liu.se 2. Supervisor: Ake Wernersson email: - 3. Institution: RAMeS, Linkoping Inst. of Tech, Linkoping, Sweden 4. Research Area: Supervisory controlled (mobile) robots 5. Summary: I'm looking at supervisory controlled robots, i.e. an operator sends commands / instructions to the remote robot which is to autonomous execute the given subtask. I have built a robot with a turnable camera and a rate gyro, and have investigated what is possible to do with this (simple) sensor combination. I will probably add a laser range scanner for autonomous wall/corridor following. Our sister group has developped algorithms for that. (Robust navigation using the Hough transform). I'm also inter in using virtual reality, but I'm not sure we have enough resources for such a project. We have my robot, and a Robosoft Robuter, some range measuring lasers, two inertial sensor systems, range cameras. The other students in my group works with: Inertial navigation, surface estimation. - Bengt Boberg Reducing ambiguites from reflective and/or transparent objects when using a laser range camera. - Jonas Nygaards Dual Control, exploratory moves, dynamic programming. - Bernt Nilsson 1. Name: Taehee Kim 2. Supervisor: Chris Malcolm 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Sensor Fusion 5. Summary: Focusing on the benefits of biological sensors and the sensor utilisation scheme, my research is aiming at implementation of a flexible control structure co-ordinating multiple sensors for assembly robots. Skin-like sensors have been developed. Application of the sensors are being investigated. 1. Name: C. Ronald Kube email: kube@cs.ualberta.ca url: http://web.cs.ualberta.ca/~kube/ 2. Supervisor: H. Zhang email: zhang@cs.ualberta.ca 3. Institution: University of Alberta, Alberta, Canada. 4. Research Area: Collective Robotics 5. Summary: This research examines the question: Can autonomous mobile robots achieve tasks collectively? We begin with the study of social insects--Nature's example of a decentralized control system--simulating those mechanisms that could prove useful in controlling teams of robots. Proposed theories are then tested on situated physical robots. To date, a system consisting of 5 mobile micro-robots have been built and used in a box-pushing task [1]. The reactive architecture is implemented in simple combinational logic, with behaviour arbitration trained using an Adaptive Logic Network (ALN) [2]. Currently, a new system of 10 micro-robots are being constructed to extend the box-pushing task to transporting [3]. Recent work has addressed the problem of stagnation recovery in reactive systems [4]. [1] Kube CR, Zhang H, (1992) "Collective Robotic Intelligence," Second International Conference on Simulation of Adaptive Behavior, 460-468. [2] Kube CR, Zhang H, Wang X, (1993) "Controlling Collective Tasks With an ALN," International Conference on Intelligent Robots and Systems IROS, 289-293. [3] Kube CR, Zhang H,(1994) "Collective Robotics: From Social Insects to Robots," Adaptive Behavior, 2(2), MIT Press, 189-219. [4] Kube CR, Zhang H, (1994) "Stagnation Recovery Behaviours for Collective Robotics," International Conference on Intelligent Robots and Systems. 1. Name: Gerard Lacey email: gerard.lacey@cs.tcd.ie 2. Supervisor: Dr. Ken Dawson-Howe email: ken.dawson-howe@cs.tcd.ie 3. Institution: Trinity College Dublin, Dublin 2, Ireland. 4. Research Area: Autonomus and Semi-autonomus Mobile Robotics 5. Summary: Developoment of a low cost multi sensor autonomus robot platfrom, intended to provide a base for further research into autonomus and semi autonomus robotic research. The future research work is focused on using exploritory moves to help remove uncertianties in the perception of the robots environment. 1. Name: David E. Lee email: dlee@cs.ucla.edu 2. Supervisors: Michel A. Melkanoff email: mam@cs.ucla.edu H. Thomas Hahn hahn@seas.ucla.edu 3. Institution: University of California, Los Angeles, CA, USA 4. Research Area: Force Control & Mating Models for Component-Component Interactions During Product Assembly Simulation 5. Summary: This research focuses on the development of force control models and representations of the dynamics of component-component interactions to predict and simulate mating conditions during product assembly. These analytic models are sought in order to provide a theoretical underpinning to virtual assembly production analysis - assessing assembly feasibility and the reliability of mating conditions prior to the physical realization of individual components and actual assembly of a product. 1. Name: David Lee email:D.Lee@cs.ucl.ac.uk 2. Supervisor: Michael Recce email: M.Recce@anat.ucl.ac.uk 3. Institution: Computer Science Department, University College London, U.K. 4. Research Area: Exploration Strategies for Mobile Robots 5. Summary: How should a mobile robot move about its environment in order to construct a high-quality world model (map) as efficiently as possible? This research addresses this question through experimentation with a sonar-equipped mobile robot. An essential sub-question is how to judge the quality of a map. A novel quality metric is defined for this purpose, using the robot's map to predict its success at executing a set of benchmark tasks. This metric is then used to examine and evaluate a number of exploration strategies which vary from purely reactive wall-following to more complex strategies which make full use of the information held in the map at each stage of the exploration. 1. Name: Wan-Yik Lee email: wylee@cs.utexas.edu URL: http://www.cs.utexas.edu/~wylee 2. Supervisor: Benjamin J. Kuipers email: kuipers@cs.utexas.edu URL: http://www.cs.utexas.edu/~kuipers 3. Institution: Artificial Intelligence Lab, URL: http://www.cs.utexas.edu/~qr/robotics.html University of Texas at Austin Department of Computer Sciences Austin, TX 78712 4. Research Area: Autonomous Mobile Robot Exploration, Mapping and Navigation using the Spatial Semantic Hierarchy (SSH) approach. Intelligent Control. 5. Summary: The Spatial Semantic Hierarchy approach to robot exploration and mapping has been developed in the context of a simulated robot, NX, and tested on simulated environments with very simple models of sensorimotor errors [Kuipers and Levitt, 1988; Kuipers and Byun, 1988, 1991]. Physical implementations of aspects of the SSH approach have been built by other researchers but they do not provide adequate demonstration of its strengths or adequate analysis of its conditions of applicability. My research will be to extend the SSH Mapping theory from its original prototypical version to a version adequate for handling real sensorimotor interaction with a real (office) environment. The extended theory will be implemented on a physical robot to explore a previously unknown environment, and to create a SSH spatial description of the environment. More in my proposal URL: http://www.cs.utexas.edu/~wylee/my-phd-proposal-abstract.html ---------------------------------------------------------------------------- 1. Name: Tsai-Yen Li email: li@flamingo.stanford.edu URL: http://robotics.stanford.edu/users/tli/bio.html 2. Supervisor: Jean-Claude Latombe URL: http://robotics.stanford.edu/users/latombe/bio.html 3. Institution: Computer Science Robotics Laboratory, Stanford University URL: http://www.stanford.edu/stanford.html 4. Research Area: On-line Robot Motion Planning 5. Summary: I'm interested in robot motion planning in general. My current research emphasize on how to bring robot motion planning on-line for dynamic environments. More specifically, I consider the scenario of a compact robotic workcell equipped with two SCARA-type manipulator arms fetching objects from a conveyer belt. The geometry of all the objects in the workspace is known in advance but everything else can be changed on-line. The problem is challenging since planning can only take very small amount of time before objects leave the workspace, and the planning time needs to be accounted for during the planning process. We approach this on-line multi-arm manipulation planning problem by decomposing the problem into four subproblems for which we developed very fast planning primitives. We have implemented our algorithm in software that simulates the robot motion and tests the on-line performance of our planner. 1. Name: Mark K. Long: long@robby.caltech.edu, long@telerobotics.jpl.nasa.gov 2. Supervisor: Joel W. Burdick: jwb@robby.caltech.edu 3. Institution: California Institute of Technology 4. Research Area: Locomotion, Sensor Based Distributed Control, Central Pattern Generators, Complex Systems...... 5. Summary: I Former Work: Kinematics and Control of Redundant Manipulators, Local/Remote Supervised Autonomy for systems with Time-Delay As member of the Technical Staff at NASA/JPL for 5 years I worked on the Kinematics and Control of Redundant Manipulators, developing Approaches for the control of the Robotics Research Arm with Composite Jacobian and Damped Least Squares Techniques. I also worked in the Supervisory Telerobotics Lab on combining Impedance Control, Generalized Compliant Motion, and Redundancy Resolution. This work included a control system for supervised autonomy with time delay, and some virtual sensing as well as distributed monitoring. II Current Work: Algorithms for Locomotion based on Central Pattern Generators and Distributed Sensor Based Control. (beginning 1993) The leg motion patterns of many 4,6,... legged animals have been shown to correspond to the stable limit cycles of coupled non-linear oscillators. Where are currently examinging this behavior as well as trying to understand robustness issues, changes in the oscillation pattern during turning motion, and the role of sensor feedback in the success of the control. Additionally, some aspects of complexity theory arise when examing the emergent behavior of the entire system of simple local controllers for each leg. It is resonable to ask: how does one design simple sensor based local controllers for each leg that when combined with the other legs through a simple central pattern generator has the emergent bahvior of stable walking and turning at a variety of speeds. 1. Name: Douglas C. MacKenzie email: doug@cc.gatech.edu Mosaic URL "file://ftp.cc.gatech.edu/pub/ai/students/doug/Doug.MacKenzie.html" 2. Supervisor: Ronald C. Arkin email: arkin@cc.gatech.edu 3. Institution: Georgia Institute of Technology, Atlanta, Georgia, USA 4. Research Area: Behavioral planning, mobile manipulation. 5. Summary: Behavior-based robot architectures are systems where the overt behavior of the system emerges from the complex interactions of numerous simple sensorimotor behaviors. The distributed nature of the overt behavior generation enormously complicates the problem of configuring the system to generate a desired overt behavior. Instead of modifying a single object, a set of sensorimotor behaviors must be selected and parameterized (a configuration) such that an appropriate overt behavior is manifested. This research will automate the process of generating a behavior configuration by creating an interactive, graphically based, configuration designer. The designer will function as an assistant, capable of pointing out areas of the design intentions which are not met by the current configuration, suggesting additions, deletions, and modifications, as well as insuring syntactic validity, semantic validity, and sufficiency of the final design. Configurations will be represented in the Configuration Description Language (CDL), a context free language which has been developed to allow compact, exact description of individual robot configurations as well as the interactions of societies of cooperating mobile robots. An optimizer module will verify that each member of the generated configuration is necessary, and also that the resulting configuration is sufficient with respect to the designer's intentions. Architecture specific code generator modules will allow generating C code using various methodologies (i.e. Subsumption, Schemas, etc.). Name: Amol Dattatraya Mali Supervisor: Amitabha Mukerjee. ( amit@iitk.ernet.in ) Institute: Centre For Robotics, Indian Institute of Technology, Kanpur, Uttar Pradesh, India, 208016. SUMMARY: We have identified stimulus overgeneralization as the cause of cyclic behaviour. We have analyzed cyclic conflict in this research. We have adopted a 3-tuple model of behaviour in which we express a behaviour module by < s, a, c> where s is stimulus, a denotes action and c denotes the consequence. We have developed a notation for power, usefulness, flexibility of behaviours. In practice to do tasks behaviours need to be triggered in a particular sequence where stimulus of each behavioural module in the chain is implied by the consequence of the module immediately preceding it. A cyclic conflict occurs when the consequence of a module later in the temporal chain of behaviours triggers its stimulus or stimulus of some module before it in the chain and the cycle is not terminated by suppression by higher level modules or by a termination condition. The cycle detection strategy that we have used is of forming a temporal graph of behaviours based on action sequences and performing graph search. The solutions that we have proposed to eliminate the cyclic conflict are (1) specialize the stimulus of module earlier in the chain so that the consequence of module later in the chain does not trigger it. (stimulus specialization) (2) Modify the action of module later in the chain so that its consequence does not trigger the stimulus of module earlier in the chain. (response generalization). 1. Name: Marinus Maris email: maris@ifi.unizh.ch URL "http://josef.ifi.unizh.ch/groups/ailab/people/maris.html" 2. Supervisor: Rolf Pfeifer email: pfeifer@ifi.unizh.ch 3. Institution: Dept. of Computer Science University of Zurich Winterthurerstrasse 190 CH - 8057 Zurich, Switzerland 4. Research Area: Autonomous Robots, Path Planning 5. Summary: To understand and design a real autonomous robot several aspects need to be investigated. The main research strategy focusses on adaptive control structures to enable the robot to manipulate its maneuvering around in the environment. As an example we have designed a robot that avoids obstacles utilizing just one sensor. 1. Name: Fred G. Martin email: fredm@media.mit.edu 2. Supervisor: Edith Ackermann email: edith@media.mit.edu 3. Institution: Media Laboratory, Mass. Inst. of Technology 4. Research Area: Robotics in Education 5. Summary: My work is concerned with the possibility of revitalizing the modern undergraduate engineering curriculum by including intensive design workshops based on the task of creating mobile autonomous robots. Included in this work is the design of hardware and software to support such activities, and the development and analysis of appropriate classroom/workshop environments. 1. Name: George Mobus email - mobus@ponder.csci.unt.edu 2. Supervisor: Paul Fisher 3. Institution: University of North Texas, Dept. of Computer Science 4. Research Area: Synthetic Brain - Artificial Neural Network Controller 5. Summary: A new learning mechanism called the Adaptrode, more closely emulates real biological synapses in encoding both activity- dependant and associative information with causal temporal ordering of conditionable to unconditionable stimuli. Embedded in an artificial neuron, and that in a network, this learning mechanism has been shown to mimic much of the classical conditioning paradigm [Mob94]. Such networks form the basis of an artificial brain in a Braitenberg vehicle [Bra84], #11, that learns to associate various non-semantic environmental cues to sensory stimuli that have meaning in the sense of reward or punishment [MF94]. The robot can also learn short-term associations which are contrary to long-term conditions without interfering with or washing out the latter [Mob94]. [Bra84] Valentino Baitenberg, "Vehicles: Experiments in Synthetic Psychology," The MIT Press, Cambridge, 1984. [Mob94] George Mobus, "Toward a theory of learning and representing causal inferences in neural networks," in D.S.Levine & M.Aparicio (Eds.) "Neural Networks for Knowledge Representation and Inference," Lawrence Erlbaum Assoc., Hillsdale NJ., 1994. [MF94] George Mobus & Paul Fisher, "MAVRIC's Brain," To be presented at Industrial & Engineering Applications of Artificial Intelligence & Expert Systems Conf., May 31 - Jun 3, 1994, Austin, TX. Proceedings to be published by Gordon and Breach Science Publishers. 1. Name: Simon P. Monckton email:monckton@mech.ubc.ca 2. Supervisor: D. Cherchas email: cherchas@cs.ualberta.ca 3. Institution: University of British Columbia, B.C., Canada. 4. Research Area: Multiagent Robotics 5. Summary: Most industrial manipulators employ a mapping between joint space and cartesian space either in the form of an inverse kinematic solution or the Jacobian inverse. This approach has evolved out of the understanding of kinematics and dynamics of mechanisms and now is the exclusive manipulator control methodology. Unfortunately, these approaches require significant support by world and dynamic models to achieve robust performance under varying environmental conditions. Furthermore, redundant manipulation often makes these approaches impractical to the point where few manufacturers consider the development of manipulators with greater than 6 d.o.f.. This research addresses a new possibility, a cooperative architecture of intelligent agents contributing toward the pursuit of a global objective while pursuing local objectives. A literature survey and early simulations indicate that this approach not only viable, but less compute intensive than existing adaptive and redundant control methods. 1. Name Jane Mulligan (mulligan@cs.ubc.ca) 2. Supervisor Alan Mackworth (mack@cs.ubc.ca) 3. Institution University of British Columbia, B.C., Canada 4. Research Area Integration of Sensing and Action 5. Summary My work looks at the sensory and model information required to achieve robotic tasks and proposes a layered structure for integrating sensing and action. Layers are organized based on the increasing informational/environmental complexity of 5 basic classes of tasks. NAME : Elizabeth Nitz enitz@mines.colorado.edu SUPERVISOR : Dr. Robin Murphy rmurphy@mines.colorado.edu INSTITUTION : Colorado School of Mines, Golden, Colorado RESEARCH AREA : Multiple Mobile Robotics & Communication SUMMARY : Current work focuses on constructing a new collaborative robot architecture consisting of one computationally-powerful "master" robot, who learns all it needs to know about a particular environment and possibly produces plans, and multiple less-powerful "apprentice" robots who then use the knowledge gathered by the master to carry out the plans or tasks within the environment. This type of system should combine the robustness of multiple homogeneous robots via redundancy with the cost-efficiency (and possibly disposability) of simple robots, and at the same time use the latest in computationally-intensive perceptual algorithms for learning. The target application is exploration and monitoring of hazardous or unfriendly environments. 1.- Name: Vicente Parra-Vega email:vega@arimotolab.t.u-tokyo.ac.jp 2.- Supervisor: Suguro Arimoto email: 3.- Institution: University of Tokyo, Tokyo, Japan 4.- Research Area: Control of Robot (adaptive/VSS/discontinuous/robustness) 5.- Summary: The research has focused on controlling robot manipulator for free (position) and constrained motion (force/position) as well by means of nonlinear techniques. Adaptive control. Discontinuous adaptive control. Variable structure control. Adaptive VSC. PD. Robustness. Stability (asymptotic and exponential). It takes into account the nonlinear model of the robot manipulator plus friction forces and the dynamics of the motor at each joint. No experiments, only theoretical work and computer simulations. 1. Name: Miles Pebody. e-mail: M.Pebody@cs.ucl.ac.uk 2. Supervisors: John Campbell. e-mail: J.Campbell@cs.ucl.ac.uk John Gilby. e-mail: 100115.624@compuserve.com 3. Institution: University College London, UK. 4. Research Area: Applied intelligent sensing and control 5. Summary: My project deals with aspects of intelligent sensing and control in a system of functionally and physically distributed control elements that are embedded and situated in a real-world environment. The system is an active laser scanning sensor device used in industry for analysing and detecting defects in products such as glass, plastic film, metal and painted surfaces which are moved through its laser beam. Reflected laser light is detected by a number of different sensors and information interpreted to locate and identify defects. This in turn can be used to direct the production process to deal with any critical situation detected. The aims of the project are to develop and explore the nature and effectiveness of the techniques used in Behaviour Based Artificial Intelligence when applied to a real world environment other than that of mobile robotics. The initial aim of the work is to develop a Subsumption Architecture based control mechanism and then to expand on initial results by exploring aspects of agent cooperation and learning. New control strategies will be experimented with which aim to increase the reliability and robustness of the system. 1. Name: Giovanni Cosimo Pettinaro 2. Supervisor: Chris Malcolm 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Behaviour Based Approach in Assembly Robots 5. Summary: Investigating the existance of a set of atomic behaviour with which describing any kind of more complex behaviour. 1. Name: Chris J. Pudney email: chrisp@cs.uwa.edu.au 2. Supervisor: Prof. Robyn Owens email: robyn@cs.uwa.edu.au 3. Institution: Univ. Western Australia, Nedlands 6009, Western AUSTRALIA 4. Research Area: Surface modelling for sensor equipped robots 5. Summary: A robot equipped with range sensors moves its sensors over the surface of an object. The sensor data obtained from the sensors is used to construct a model of the surface, and the surface is used in turn to control the robot's motion. Thus the surface model is constructed on-line. 3-D surface modelling techniques and algorithms for controlling the robot's surface following motion are being developed. Name : Ranganathan Ramanathan (AKA Rungun) email : ramanath@asel.udel.edu Supervisor : Dr. Rahmim Seliktar & email : seliktr@duvm.ocs.drexel.edu Dr. Tariq Rahman email : rahman@asel.udel.edu Institution : Drexel University, MEM department, Philadelphia, PA 19014, USA Research Area : Rehabilitation Robotics Summary : Design and development of powered orthosis. Stuck at an interesting but tough problem of finding out an GOOD anti-gravity mechanism to use. Then we power this mechanism, and look into various control issues and human machine interface. 1. Name: Dan S Reznik email: reznik@robios.me.wisc.edu 2. Supervisor: Vladimir Lumelsky email: lumelsky@robios.me.wisc.edu 3. Institution: University of Wisconsin-Madison, Madison, WI, 53706, USA 4. Research Area: Sensor-based motion planning for highly-redundant kinematic structures 5. Summary: I am working on the design of sensor-based algorithms for for highly-redundant robots -- so far I have considered snake-shaped robots, multi-finger hand with lots of links per finger, and "multi-branch" snakes, which are tree-shaped robots with lots of degrees of freedom. We consider both planar and 3D structures. 1. Name: Julio Kenneth Rosenblatt email: jkr@ri.cmu.edu 2. Supervisor: Chuck Thorpe email: cet@ri.cmu.edu 3. Institution: Robotics Institute, Carnegie Mellon University Pittsburgh, PA, USA 4. Research Area: Mobile Robot Architectures 5. Summary: The Distributed Architecture for Mobile Navigation (DAMN) provides a framework for independent, distributed, task-achieving behaviors, similar in spirit to the Subsumption Architecture. One important difference between DAMN and the Subsumption Architecture is that rather than one behavior overriding another, DAMN behaviors send weighted votes to an arbiter wheich then selects the action that best satisfies several objectives concurrently. 1. Name: Peter Scheffel email: peter@concave.cs.wits.ac.za 2. Supervisor: Conrad Mueller email: conrad@concave.cs.wits.ac.za 3. Institution: University of the Witwatersrand, Johannesburg, SA 4. Research Area: Path planning for manipulators with DOF<6 5. Summary: The aim of the research is to find genrally applicable techniques to improve the performance of path planning without precomputing the configuration space. Initially an implementation of an approach that does precompute the configuration space was attempted. This was found to have very poor performance especilly on simple cases like an empty environment! The research has therefore concentrated on finding solutions to simple problems quickly. Results achieved so for obvious paths in 3 DOF for a 2D env. take under 20 seconds, on a standard 486 PC, with some paths only taking 3 seconds. Most paths can be found in under 10 minutes and memory limitations hinder paths that could take more than an hour. Optimisations have played a large part in the feasiblity of the reasearch. Improvements of the order of 5 fold are not uncommon using compulational geometry techniques to improve the geometric intersections of lines. 1. Name: Jeff Schneider email: schneider@cs.rochester.edu 2. Supervisor: Chris Brown email: brown@cs.rochester.edu 3. Institution: University of Rochester 4. Research Area: Acquisition of Robot Motor Skills 5. Summary: In open loop skills such as throwing a ball, an entire robot control sequence can be viewed as a single point in a high dimensional space. Then, the problem of improving accuracy as well as increasing range of performance is a search problem. We have implemented a throwing robot with a flexible link. We have designed an efficient way to search the space with the result that our robot learns to "whip" its flexible link at the right frequency to produce long throws. The result is particularly encouraging since the robot was not given any model of its flexible link, and no samples of the "whipping" motion were ever shown to it. Recent work considers the acquisition and improvement of closed loop skills. Highly skilled humans have the ability to perform complex motions relatively open loop (consider a hockey player that corners in a single smooth motion compared to the beginner that must concentrate on balance throughout the turn). We believe that closed loop skill acquisition can benefit from an attempt to make the skills more open loop as learning progresses. 1. Name: Armin Sulzmann email: sulzmann@imtsg1.epfl.ch 2. Supervisor: R.Clavel email: 3. Institution: Swiss Federal Institute of Technology, Lausanne, Switzerland 4. Research Area: Micro-Robotics 5. Summary: This research examines the question: Developement of a vision-based (virtuel-reality) System to guide the manipulations of microsystems, microstructurs, etc. 1. Name: Hans Tangelder email:j.w.h.tangelder@io.tudelft.nl 2. Supervisor: Joris Vergeest 3. Institution: Delft University of Technology Faculty of Industrial Design Engineering Delft, The Netherlands. 4. Research Area: Rapid Prototyping using Robot Milling 5. Summary: One of the favourable techniques for rapid shape prototyping is numerically controlled milling. For this purpose we have installed a Sculpturing Robot work cell consisting of a 6-DOF industrial robot and a turn table on which a foam stock is placed. The Sculpturing Robot work cell is driven by an off-line generated path file. This path is calculated from the data representing the shape, or object, that must be replicated. This path is the result of a strategy that must take into account the limits of tool work space as well as the shape and placement of all obstacles within this space. The proposed project aims at finding appropriate methods for such a process. They must meet severe requirements concerning safety, autonomy, robustness, accuracy, speed and insensitivity to imperfections of the input geometry. 1. Name: Eddie Tunstel email: tunstel@chama.eece.unm.edu or tunstel@robotics.jpl.nasa.gov 2. Supervisor: Dr. M Jamshidi email: jamshid@houdini.eece.unm.edu 3. Institution: University of New Mexico, Albuquerque 4. Res Area: Fuzzy and Intelligent Control of Mobile Robots 5. Summary: This research focusses on the development of hybrid intelligent control architectures for autonomous mobile robots and mobile manipulation. The work includes investigations of various combinations of paradigms such as fuzzy logic, neural networks, behavior control, and genetic algorithms for real time motion control. The research focus is on control architectures for navigation, path planning, and environment mapping with empahasis on embedded application. 1. Name: Cem Unsal email: unsal@blackbox.cl.ee.vt.edu 2. Supervisor: John S. Bay email: bayj@vtvm1.cc.vt.edu 3. Institution: Virginia Tech, Blacksburg, VA, USA 4. Research Area: Multiple Mobile Robots (Army-ant Project) 5. Summary: My research is based on the idea of using a large homogeneous population of mobile robots as a transportation system. Army-ant scenario may also be applied to space/underwater missions. We treat "army-ant swarm" as a self-organizing system. Robots are simple in terms of knowledge and/or communication abilities. Important characteristic of the scenario are: lack of map knowledge, large number of agent, non-hiererchical structure, emergence (of some form) of intelligence from local interactions and simple behavioral rules. I'm currently working on behavioral self-organization (decision systems) of multiple agents. 1. Name: Jim Vaughan email: jev1@uk.ac.bton.unix 2. Supervisor: Graeme Awcock email: gja@unix.bton.ac.uk 3. Institution: University of Brighton, UK 4. Research Area: Machine vision/sensor fusion 5. Summary: Using intelligent sensors, particularly vision, to generate an accurate perception of a machine's environment, to allow it to operate with a high degree of reliability 1. Name: Virgilio B. Velasco Jr., aka "Dean" email: vbv@pris.eeap.cwru.edu 2. Supervisor: Wyatt S. Newman email: wsn@pris.eeap.cwru.edu 3. Institution: Electrical Engineering and Applied Physics Department Case Western Reserve University AND Center for Automation and Intelligent Systems Research 4. Research Area: Agile Manipulation 5. Summary: Agile manufacturing is a revolutionary product assembly approach which allows the manufacturing of a wide variety of products, with selectable features and batch sizes. For a robot to do highly agile manufacturing, it must also be agile in its ability to grasp a wide variety of objects. This research project will focus on techniques for grasping and manipulating many different objects with a minimal selection of grippers and external fixtures. It will also seek to grasp such objects within a minimal amount of time. 1. Name: Richard Voyles email: robodude@cmu.edu URL: http://www.cs.cmu.edu:8001/afs/cs.cmu.edu/user/deadslug/ftp/home.html 2. Supervisor: Pradeep Khosla email: pkk@ri.cmu.edu 3. Institution: Carnegie Mellon University, Pittsburgh, PA, USA URL: http://www.cs.cmu.edu:8001/afs/cs.cmu.edu/user/mwgertz/www/aml.html 4. Research Area: Multi-Agent Control/Perception 5. Summary: I'm investigating the cooperation of relatively dumb agents with minimal communication channels during control and perception tasks. I'm applying systems of encapsulated agents to control of a Utah/MIT dextrous hand, control of a Puma 560, and possbily to the task of selecting control methodologies for a robot. 1. Name: Gabriel D. Warshaw email: gabriel@sce.carleton.ca 2. Supervisor: Howard Schwartz 3. Institution: Carleton University, Ottawa, Ontario, Canada 4. Research Area: Sampled-Data Robot Adaptive Control 5. Summary: I am addressing the stability and performance of discretized adaptive control algorithms for robotic manipulator control, and the compensation of these algorithms for improved stability and tracking performance. The discretization of adaptive control algorithms published in the literature can result in a sampled-data robot system for which stability has not been guaranteed. By formulating the entire sampled-data system in continuous-time, I have used Lyapunov's direct method to determine the stability and to derive a non-linear discrete-time compensating term. I have demonstrated the theoretical results through simulation and implementation on a 2 degree-of-freedom direct drive manipulator. 1. Name: Martin D. Westhead 2. Supervisor: John Hallam 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Theoretical underpinnings of behaviour based systems 5. Summary: This work is still in its early stages, but will attempt to apply formalisms such as process algebra's, dynamical systems and petri nets to the problem of behaviour based robotic control. The goal of the work is a better understanding of the parallel interaction of behaviour based systems with the hope that this might provide the foundation for a more rigorous design methodology than those currently employed. 1. Name: Jeremy Wyatt 2. Supervisor: Gillian Hayes and John Hallam 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Learning in Mobile Robots 5. Summary: I an interested in using learning to help design robot controllers. I am working on combining a technique called reinforcement learning with a behaviour based controller. Elementary behaviours are learned separately, and then coordinating behaviours are learned. Ultimately the aim is to build a hierarchy of behaviours with several levels. 1. Name: Gordon Wyeth email: wyeth@elec.uq.oz.au 2. Supervisor: Mark Schulz email: marks@elec.uq.oz.au 3. Institution: Robotics Laboratory, Dept. of Electrical and Computer Engineering, University of Queensland, Brisbane, Australia. 4. Research Areas: Artificial Neural Control of a Hunt and Gather Robot, Micromice. 5. Summary: Hunt and Gather Robot The main aim of this project is to produce models of cognitive structures that support intelligent behaviour sufficient to allow a mobile robot to perform collecting tasks. The project involves the construction of a robot dog, CORGI, that retrieves tennis balls around the lab. The techniques used to build this robot are similar to those used in behaviour-based robotics, but are based on a neural paradigm. The robot's primary sense is a CCD camera that is used to locate the tennis balls, sense obstacles and to provide visual cues to allow the robot to return to its home position. The artificial neural network architecture is based on a combination of conventional networks (MLP, SOM) and constructs found in Braitenberg vehicles. Micromice Micromice are autonomous maze solving robots that are entered in competitions to see who can solve the maze and run the fastest path in the least time. My Micromice include CUQEE I & II, Australian Micromouse Champions. CUQEE III is currently in development and should debut at the end of 1994. 1. Name: Brian Yamauchi email: yamauchi@alpha.ces.cwru.edu 2. Supervisor: Randall Beer email: beer@alpha.ces.cwru.edu 3. Institution: Department of Computer Engineering and Science Case Western Reserve University, Cleveland, OH 4. Research Area: Mobile Robots/Neural Networks/Genetic Algorithms 5. Summary: My research involves the use of genetic algorithms to evolve recurrent neural networks for the control of autonomous agents and mobile robots. My previous work concentrated on generating networks that can generate sequential behavior and learn from environmental reinforcement. Current research is focused on landmark-based navigation and learning tasks in simulated environments. I have also applied these techniques to evolve controllers for predator avoidance and landmark recognition on a real Nomad 200 mobile robot equipped with sonar sensors (at the Navy Center for Applied Research in AI at the Naval Research Laboratory). My long-term goal is to put these pieces together to develop a mobile robot system capable of navigation and spatial learning in dynamic real world environments. 1. Name: Mark Yim email: mark@killdeer.stanford.edu 2. Supervisor: J.C. Latombe email: latombe@cs.stanford.edu 3. Institution: Stanford University, Stanford CA, 94305 4. Research Area: Reconfigurable Modular Robot Locomotion 5. Summary: A dynamically reconfigurable modular robot named Polypod has been designed, simulated and partially constructed. Research is being done on unusual statically stable locomotion gaits implemented on Polypod, for example, a rolling loop, a moving carpet with many feet, slinky locomotion... Each gait is achieved with a very simple behaviour based control scheme. A taxonomy of locomotion and the kinematics of locomotion will be analyzed. Article 11554 of comp.robotics: Xref: glinda.oz.cs.cmu.edu comp.robotics:11554 Path: honeydew.srv.cs.cmu.edu!nntp.club.cc.cmu.edu!godot.cc.duq.edu!newsfeed.pitt.edu!gatech!newsxfer.itd.umich.edu!nntp.cs.ubc.ca!alberta!kube From: kube@cs.ualberta.ca (Ron Kube) Newsgroups: comp.robotics Subject: July's Grad. Students Who's Who in Robotics Date: 5 Jul 1994 15:10:22 GMT Organization: Computing Science, U of Alberta, Edmonton, Canada Lines: 1117 Message-ID: <2vbt4u$8q4@scapa.cs.ualberta.ca> NNTP-Posting-Host: kitscoty.cs.ualberta.ca Summary: Monthly posting July Keywords: who's who in robotics >>>>>>>>>>>>>> GRAD STUDENTS WHO'S WHO IN ROBOTICS <<<<<<<<<<<<<< ================================================================= Have you ever wondered what grad students are doing in robotics? A trip to your local research library allows you to see Who's Who in robotics at the post-doc level, i.e. Professor SoNso, and SuchNsuch, but what about the graduate students working on their MSc. or PhD? Here is a summary of the received entries to date. If you would like to appear in the Grad Students Who's Who in Robotics send a note to kube@cs.ualberta.ca using the 5 point format. If you have WWW home-page then include its URL after your name. This file is available via anonymous ftp from ftp.cs.ualberta.ca in directory pub/kube as file whosWho, and as a WWW web page courtesy of Johan Forsberg (jf@sm.luth.se). =============================================================================== 1. Name: Arvin Agah email: arvin@robotics.usc.edu 2. Supervisor: George A. Bekey email: bekey@robotics.usc.edu 3. Institution: University of Southern California Institute for Robotics & Intelligent Systems 4. Research Area: Multi-Robot Systems & Biologically-Inspired Robotics 5. Summary: Investigating the issues of group behavior in robot colonies, both in simulation and hardware 1. Name: Peter O Aberg email: tm90pa@hh.se Anders Borghed email: tm90ab@hh.se 2. Supervisor: Per-Arne Wiberg email: pelle@hh.se 3. Institution: Centre for Computer Architecture, Halmstad University, Sweden 4. Research Area:Future Robotic Control 5. Summary: The direct and inverse kinematic problems are fundamental issues in robot modelling and control. The direct kinematics is a mapping which allows the position of the robot in the cartesian space to be related with the position in the joint space, i.e. given the position in the joint coordinates, we can univocally derive, by means of the direct kinematics, the corresponding position in the cartesian coordinates. The kinematic problem is different depending on the structure of the robot. For open-chain robotic structures the inverse kinematic poses the most difficulties and for closed-chain robotic structures it's the opposite. There are many different techniques for solving the kinematic problems. Most methods used today are based upon traditional mathematics, such as Newton-Raphsons method. None of these methods are in any way perfect for the problem. We have analyzed several different mechanical structures and control strategies that are in use today. In our work we also present methods to improve robotic motion performance by using new control methods and new mechanical structures. 1. Name: Karl R Altenburg email: altenbur@plains.nodak.edu 2. Supervisor: Mark Pavicic email: pavicic@plains.nodak.edu 3. Institution: North Dakota State University, Fargo, ND, USA 4. Research Area: Multiple Mobile Robots 5. Summary: Investigating the efficiency gains provided by communication and memory during multirobot search and retrieval type tasks. Currently tests are being conducted on a set of six small mobile robots, and in simulation. The work also investigates reactive control for individual robots and emergent control for the system. 1. Name : Venkateswara Rao Ayyadevara (email : avrao@vax2.concordia.ca) 2. Supervisors : Dr.R.M.H. Cheng (email : richard@vax2.concordia.ca) Dr.Ramesh Rajagopalan (email : ramesh@vax2.concordia.ca) 3. Institution : Centre for Industrial Control, Dept. of Mechanical Engineering Concordia University Montreal, Canada 4. Research Topic : Development of an Automated Robotic Deburring Workcell for Impeller Blades 5. Summary : Impeller is a component used in aircraft engines. Owing to their geometrical complexity and the rigorous standards specified by air safety regulations, impellers are extremely expensive to produce. After several thousand hours in use, the blades of an impeller are warped and the edges are corroded. If some of these impeller blades can be refurbished after being used for several thousand hours, considerable amount of money could be saved. I am working as part of a team which is developing an automated workcell using Yamaha Zeta Deburring robot to probe the surface of the impeller blade, reconstruct the surface, determine the desired edge profile and then use the robot to machine the workpiece to obtain that edge profile. My task is to develop set up for probing the surface of the impeller blades and to interface the controller of the robot with a PC-transputer network which is responsible for directing the probe, surface reconstruction and control of the robot. 1. Name: Tucker Balch email: tucker@cc.gatech.edu 2. Supervisor Ronald Arkin email: arkin@cc.gatech.edu 3. Institution Georgia Tech 4. Research Area: Communication in Autonomous Robot Societies 5. Summary: Multiple cooperating robots are able to complete many tasks more quickly and reliably than one robot alone. Communication between the robots can multiply their capabilities and effectiveness, but to what extent? In our research, the importance of communication in reactive robotic societies is investigated through experiments on both simulated and real robots. So far, our research has shown that for some tasks communication can significantly improve performance, but for others inter-agent communication is apparently unnecessary. In cases where communication helps, the lowest level of communication is almost as effective as the more complex type. Research is being extended to more complex scouting tasks for the Army. 1. Name: Johan G Benade email: jgb@ing1.rau.ac.za 2. Supervisor: Andre L Nel email: aln@ing1.rau.ac.za 3. Institution: Rand Afrikaans University, Johannesburg, RSA. 4. Research Area: Autonomous Robotics 5. Summary: Research is aimed at producing an improved biologically based controller for use in hexapod locomotion. The leg controller must be able to cope with uneven terrain - gaps in surfaces - inclines and surface tension variability. At the end of the project a functioning hardware realisation must be produced. 1. Name: Todd M. Bezenek email: bezenek@plains.nodak.edu 2. Super: Mark Pavicic email: pavicic@plains.nodak.edu 3. Istit: North Dakota State University, Fargo, ND 4. Area: Communications for multiple, autonomous robots. 5. Summary: Several groups are working with multiple robots to collectively solve a single problem. Those addressing the problem of communication between robots are assuming that there exists an unbreakable data path between each pair of robots, or between each robot and a central station. In many real applications where multiple robots may be used, communication between each pair of robots may not be continuous. As the robots move, the network representing pairs of robots that are able to successfully communicate changes. My goal is to develop a protocol which will allow robots on this network to communicate effectively. I have built two robots which communicate at 1200 baud over a simplex 49Mhz data channel. A third, which will act as a slave attached to a PC, is currently being constructed. 1. Name: William Chesters (http://www.dai.ed.ac.uk/students/williamc/williamc.html) 2. Supervisors: Gillian Hayes, John Hallam 3. Institution: Department of AI, University of Edinburgh 4. Research Area: Robot learning with neural networks 5. Summary: Neural nets look like an interesting approach to `bottom-up' AI, but if you try to apply them to non-trivial robot tasks, you come up against some serious problems: - a net tends to forget old knowledge as new experiences come in - it tends to get stuck in `local minima' - it can only work at one timescale: it can't support hierarchies of behaviours I'm interested in getting round these problems by using a community of competing nets. 1. Name: Howie Choset email: choset@robby.caltech.edu 2. Supervisor: Joel W. Burdick email: jwb@robby.caltech.edu 3. Institution: California Institute of Technology 4. Research Area: Sensor Based Planning for Mobile and Hyper-redundant Robots. 5. Summary: ``Sensor Based Planning'' incorporates sensor information, reflecting the current state of the environment, into a robot's planning process, as opposed to classical planning, which assumes full knowledge of the world's geometry prior to planning. Sensor based planning is important because: (1) the robot often has no a priori knowledge of the world; (2) the robot may have only a coarse knowledge of the world because of limited memory; (3) the world model is bound to contain inaccuracies which can be overcome with sensor based planning strategies; and (4) the world is subject to unexpected occurrences or rapidly changing situations. Currently, we are working on some initial steps towards path planning in a static environment where there is no a priori knowledge. We are develop- ing an incremental method to construct a Generalized Voronoi Graph (GVG), which is a 1-dimensional retract of a bounded space. The GVG is the same thing as a Generalized Voronoi Diagram in two dimensions. Like many other path planning schemes, the distance function is an integral part of the GVG. This function is nonsmooth; it is shown that the non-smoothness occurs at points which are ``critical'' to many path planning schemes. We have done some nonsmooth analysis on the distance function which has lead to the incorporation of simple and realistic sensor models. 1. Name: Chris Connolly email: connolly@cs.umass.edu 2. Supervisor: Rod Grupen 3. Institution: Laboratory for Perceptual Robotics, University of Massachusetts, Amherst, MA (USA) 4. Research Area: Motor and task planning using harmonic functions 5. Summary: Harmonic functions are solutions to Laplace's equation, and can be rapidly computed using resistive networks. They exhibit no local minima, and can be used to generate smooth goal-reaching trajectories [1]. We're using them for coarse reaching (on P-50 hand/arm systems) and mobile robot trajectory planning (on an unmanned ground vehicle). The resistive network formulation also turns out to be useful for modeling certain nuclei of the basal ganglia [2,3], and provide a theory for aspects of motor and cognitive planning in the mammalian central nervous system. [1] Connolly CI, Grupen RA, (1993) "The Applications of Harmonic Functions to Robotics", Journal of Robotic Systems, 10(7):931-946. [2] Connolly CI, Burns JB, (1993) "A Model for the Functioning of the Striatum", Biological Cybernetics, 68(6):535-544. [3] Connolly CI, Burns JB, (1993) "A New Striatal Model and its Relationship to Basal Ganglia Diseases", Neuroscience Research, 16:271-274. 1. Name: Joe Cronin J.Cronin@UNSW.edu.au 2. Supervisor: Richard Frost Richard Wilgoss 3. Institution: University of New South Wales, Sydney, Australia. 4. Research Area: Biped Robot. 5. Summary. I'll make it brief. It will have two legs. It will be anthropomorphic. It will walk. The ultimate goal of this project is to design, model and build a biped robot platform, capable of dynamic motion. It will be hydraulically driven, use an HC11 on every joint and stand about four feet high. There are two main areas of research; the first is to develop and control an ankle and foot with all degrees of freedom of the human ankle and foot, the second is to use distributed HC11's to solve the inverse kinematics at the joint. The project is at the stage where construction will begin before the end of 1993. As the school is in some financial difficulty, I have had to raise all funds for this project privately. The school does not have an active mobile robot group, I would be interested in anyone who would be interested in me. 1. Name: Bruce Digney digney@dvinci.usask.ca 2. Supervisor: M. M. Gupta 3. Institution: Dept. of Mech. Eng., University of Saskatchewan (Canada) 4. Research Area: Distributed Adaptive Control Systems 5. Summary: In my research I propose that by incorporated learning and adaption into a behavior based control system, the skills and behaviors which are impossible or impractical to be predetermined and embedded can be learned by the robot during operation. The result is a distributed adaptive control system (DACS), which can be thought of as the robot's artificial adaptive nervous system. This DACS autonomously learns the sensory-response couplings between the highest behavioral level, where the desired tasks are specified, and the lowest level actuators, which ultimately perform those tasks. A DACS has been developed for a simulated quadruped mobile robot and extensions to a physical robot are planned. 1. Name: Hans Dulimarta email: dulimart@cps.msu.edu 2. Supervisor: Anil K. Jain email: jain@cps.msu.edu 3. Institution: PRIP Laboratory Department of Computer Science Michigan State University East Lansing, Michigan 48824 4. Research Area: Distributed Robotics 5. Summary: I address the problem of task decomposition in a mobile robot navigation system. The underlying supposition of my approach is that in a typical robot navigation system, there are a number of modules running concurrently and each module is assigned a specific subtask. In order to accomplish the common goal of the navigation task, these modules share the resources and common data in the system. In such a system, resource access control and information sharing among the modules must be properly managed. In this work, I propose a decomposition of a robot navigation system into a number of {\it client} and {\it server} modules. Resource access control, resource sharing, information sharing, and process synchronization in the entire robot navigation system are delegated to the server modules. The clients send appropriate requests in order to avail of these facilities. There are two types of server modules defined in the system: {\it data server} and {\it hardware server}. The core of the system consists of one data server and several hardware servers. The data server acts as a common information exchange medium for all the clients, while the hardware servers provide access interface to the hardware or peripherals on the robot. By decoupling the hardware access routines from the hardware servers, the modules in the navigation system can be made independent of hardware platform being used. 1. Name: Sean P. Engelson email: engelson@cs.uchicago.edu 2. Supervisor: Drew V. McDermott email: mcdermott@cs.yale.edu 3. Institution: Yale University 4. Research Area: Map Learning 5. Summary: My work explores a `passive' mapping paradigm, in which the map-learning system has no direct control over the agent's actions. The main problem in map-learning is the fact that the agent's location is never perfectly known. Errors in localization lead inevitably to mapping errors. Passive mapping exacerbates this problem, since the mapper cannot perform experiments to verify the robot's location. My approach allows mapping errors to occur, and deals with them in two ways. First, is the use of a graph-based representation scheme which incorporates both connectivity and positional information to locally bound mapping error. Second, errors are diagnosed and repaired as information becomes available. The diagnosis and repair strategies are based on a taxonomy of possible mapping errors. 1. Name: Ted C. Feltmeyer email-tedhead@csd4.csd.uwm.edu email-felt1512@watt.cae.uwm.edu 2. Supervisor: Robert Borchelt email-borchelt@convex.csd.uwm.edu (Dr. Bob) 3. Institution: University of Wisconsin Milwaukee, Milwaukee, WI 4. Research Area: Automated diagnosis and error recovery in robotic workcells using artificial intelligence methods 5. Summary: I have worked since August 93 setting up a functional robotic workcell with one robot acting as a slave to the other. At this stage, the cell will perhaps perform some sort of electronic assembly operations. This physical system will become a testbed for different AI control techniques. My thesis will come out of the physical system setup and initial control using a hybrid expert system. My dissertation will expand further the different control techniques, plus integrate vision technology. The physical system is based on two Adeptone robots, AB PLC2/30, and (with luck) a new Pentium based PC. 1. Name: Johan Forsberg email: jf@sm.luth.se 2. Supervisor: Ake Wernersson 3. Institution: Robotics & Automation Lulea University of Technology S-971 87 LULEA, SWEDEN 4. Research Area: Autonomous Robots 5. Summary: I'm currently working with statistical map representation where the robot builds the map autonomously from measurements taken by a scanning range measuring laser. This statistical map is mainly intended for localization, while some other approach might be better for path planning. I think that to create an autonomous robot, we should not try to mimick the way humans, or animals, work. A better approach is to look at what computers are good at, and how this can be used to make a better robot. One example: when entering a room, we can know which room we are in either by recognizing the way the room looks, or by knowing exactly what route we took to get there (in meters and millimeters). The first method is the main method for humans, but extremely difficult for a computer. The second is difficult for a human, but easy for a robot. 1. Name: Bill Gribble email: grib@mamba.asg.arlut.utexas.edu 2. Supervisor: Ben Kuipers email: kuipers@cs.utexas.edu 3. Institution: University of Texas AI Lab 4. Research area: Distributed architectures for visually guided mobile robots 5. Summary: A robotic platform capable of manipulating increasingly large amounts of sensory data while still meeting real-time performance constraints must diverge from a monolithic model of computation. With a moderate number of compute nodes, each specialized for a task, the problem of sensor data partitioning can be more easily attacked. We are constructing a robot consisting of a number of nodes, including general-purpose microprocessors, small microcontrollers, and DSP engines, each with its own purpose, all linked by a 32Mbps custom local network. The software architecture is based on a packet protocol similar to that used in dataflow parallel machines. 1. Name: Lyle Hall email: hall@cs.uiuc.edu 2. Supervisor: Sylvian Ray email: ray@cs.uiuc.edu Fred Delcomyn delcomyn@ux1.cso.uiuc.edu 3. Institution: University of Illinois at Urbana-Champaign 4. Research Area: Robot Simulation and Control of Insect-like Walking 5. Summary: Developing a dynamic simulator and controller to model walking in cockroaches. 1. Name: Bridget Hallam 2. Supervisor: Gillian Hayes 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Controlling Robots using Biological Theories 5. Summary: Studying animal behavioural control can give insights into autonomous behaviour that may prove useful for those wishing to build autonomous robots. Implementing Halperin's neuro-connector model of learning and motivation on a mobile robot has shown that it can be used to control robots, and that it is reasonably complete. Implementation in simulation will discover the sensitivity of the various features to variations in parameters and the exact equations used, and so improve the model as a robot controller. It may also improve the model for ethologists. 1. Name: Roger B. Hertz (hertz@ecf.toronto.edu) 2. Supervisor: Peter C. Hughes (hughesp@ecf.toronto.edu) 3. Institution: University of Toronto Institute for Aerospace Studies 4. Research Area: Articulated-Truss Manipulators 5. Summary: We are investigating the use of articulated truss mechanisms for both space and terrestrial robotics applications. We have constructed a prototype manipulator based on this concept that is capable of 3-DOF spatial motion. My research is centered on applying the technology to a 6-DOF industrial version of the manipulator. Current work is involved with manipulator design, development of kinematics algorithms, workspace analysis, and customization of an industrial robot contoller. 1. Name: Tomas Hogstrom email: tomas@idefix.ikp.liu.se 2. Supervisor: Ake Wernersson email: - 3. Institution: RAMeS, Linkoping Inst. of Tech, Linkoping, Sweden 4. Research Area: Supervisory controlled (mobile) robots 5. Summary: I'm looking at supervisory controlled robots, i.e. an operator sends commands / instructions to the remote robot which is to autonomous execute the given subtask. I have built a robot with a turnable camera and a rate gyro, and have investigated what is possible to do with this (simple) sensor combination. I will probably add a laser range scanner for autonomous wall/corridor following. Our sister group has developped algorithms for that. (Robust navigation using the Hough transform). I'm also inter in using virtual reality, but I'm not sure we have enough resources for such a project. We have my robot, and a Robosoft Robuter, some range measuring lasers, two inertial sensor systems, range cameras. The other students in my group works with: Inertial navigation, surface estimation. - Bengt Boberg Reducing ambiguites from reflective and/or transparent objects when using a laser range camera. - Jonas Nygaards Dual Control, exploratory moves, dynamic programming. - Bernt Nilsson 1. Name: Juergen Karner email: juergen@lpa.uni-sb.de 2. Supervisor: H. Janocha 3. Institution: University of Saarland Institute for Process Automation 4. Research Area: Hybrid Control of Robots with Fuzzy Logic 5. Summary: Existing robots use PID-controllers. The parameters of the PID-controllers are fixed and do not change depending on the actual pose of the robot. We want to change the parameters of the PID-controllers using fuzzy-logic. 1. Name: Taehee Kim 2. Supervisor: Chris Malcolm 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Sensor Fusion 5. Summary: Focusing on the benefits of biological sensors and the sensor utilisation scheme, my research is aiming at implementation of a flexible control structure co-ordinating multiple sensors for assembly robots. Skin-like sensors have been developed. Application of the sensors are being investigated. 1. Name: C. Ronald Kube email: kube@cs.ualberta.ca url: http://web.cs.ualberta.ca/~kube/ 2. Supervisor: H. Zhang email: zhang@cs.ualberta.ca 3. Institution: University of Alberta, Alberta, Canada. 4. Research Area: Collective Robotics 5. Summary: This research examines the question: Can autonomous mobile robots achieve tasks collectively? We begin with the study of social insects--Nature's example of a decentralized control system--simulating those mechanisms that could prove useful in controlling teams of robots. Proposed theories are then tested on situated physical robots. To date, a system consisting of 5 mobile micro-robots have been built and used in a box-pushing task [1]. The reactive architecture is implemented in simple combinational logic, with behaviour arbitration trained using an Adaptive Logic Network (ALN) [2]. Currently, a new system of 10 micro-robots are being constructed to extend the box-pushing task to transporting [3]. Recent work has addressed the problem of stagnation recovery in reactive systems [4]. [1] Kube CR, Zhang H, (1992) "Collective Robotic Intelligence," Second International Conference on Simulation of Adaptive Behavior, 460-468. [2] Kube CR, Zhang H, Wang X, (1993) "Controlling Collective Tasks With an ALN," International Conference on Intelligent Robots and Systems IROS, 289-293. [3] Kube CR, Zhang H,(1994) "Collective Robotics: From Social Insects to Robots," Adaptive Behavior, 2(2), MIT Press, 189-219. [4] Kube CR, Zhang H, (1994) "Stagnation Recovery Behaviours for Collective Robotics," International Conference on Intelligent Robots and Systems. 1. Name: Gerard Lacey email: gerard.lacey@cs.tcd.ie url: http://cvg.cs.tcd.ie/gjlacey/doc/personal/personal.html 2. Supervisor: Dr. Ken Dawson-Howe email: ken.dawson-howe@cs.tcd.ie 3. Institution: Trinity College Dublin, Dublin 2, Ireland. 4. Research Area: Autonomus and Semi-autonomus Mobile Robotics 5. Summary: Developoment of a low cost multi sensor autonomus robot platfrom, intended to provide a base for further research into autonomus and semi autonomus robotic research. The future research work is focused on using exploritory moves to help remove uncertianties in the perception of the robots environment. 1. Name: David E. Lee email: dlee@cs.ucla.edu 2. Supervisors: Michel A. Melkanoff email: mam@cs.ucla.edu H. Thomas Hahn hahn@seas.ucla.edu 3. Institution: University of California, Los Angeles, CA, USA 4. Research Area: Force Control & Mating Models for Component-Component Interactions During Product Assembly Simulation 5. Summary: This research focuses on the development of force control models and representations of the dynamics of component-component interactions to predict and simulate mating conditions during product assembly. These analytic models are sought in order to provide a theoretical underpinning to virtual assembly production analysis - assessing assembly feasibility and the reliability of mating conditions prior to the physical realization of individual components and actual assembly of a product. 1. Name: David Lee email:D.Lee@cs.ucl.ac.uk 2. Supervisor: Michael Recce email: M.Recce@anat.ucl.ac.uk 3. Institution: Computer Science Department, University College London, U.K. 4. Research Area: Exploration Strategies for Mobile Robots 5. Summary: How should a mobile robot move about its environment in order to construct a high-quality world model (map) as efficiently as possible? This research addresses this question through experimentation with a sonar-equipped mobile robot. An essential sub-question is how to judge the quality of a map. A novel quality metric is defined for this purpose, using the robot's map to predict its success at executing a set of benchmark tasks. This metric is then used to examine and evaluate a number of exploration strategies which vary from purely reactive wall-following to more complex strategies which make full use of the information held in the map at each stage of the exploration. 1. Name: Wan-Yik Lee email: wylee@cs.utexas.edu URL: http://www.cs.utexas.edu/~wylee 2. Supervisor: Benjamin J. Kuipers email: kuipers@cs.utexas.edu URL: http://www.cs.utexas.edu/~kuipers 3. Institution: Artificial Intelligence Lab, URL: http://www.cs.utexas.edu/~qr/robotics.html University of Texas at Austin Department of Computer Sciences Austin, TX 78712 4. Research Area: Autonomous Mobile Robot Exploration, Mapping and Navigation using the Spatial Semantic Hierarchy (SSH) approach. Intelligent Control. 5. Summary: The Spatial Semantic Hierarchy approach to robot exploration and mapping has been developed in the context of a simulated robot, NX, and tested on simulated environments with very simple models of sensorimotor errors [Kuipers and Levitt, 1988; Kuipers and Byun, 1988, 1991]. Physical implementations of aspects of the SSH approach have been built by other researchers but they do not provide adequate demonstration of its strengths or adequate analysis of its conditions of applicability. My research will be to extend the SSH Mapping theory from its original prototypical version to a version adequate for handling real sensorimotor interaction with a real (office) environment. The extended theory will be implemented on a physical robot to explore a previously unknown environment, and to create a SSH spatial description of the environment. More in my proposal URL: http://www.cs.utexas.edu/~wylee/my-phd-proposal-abstract.html ---------------------------------------------------------------------------- 1. Name: Tsai-Yen Li email: li@flamingo.stanford.edu URL: http://robotics.stanford.edu/users/tli/bio.html 2. Supervisor: Jean-Claude Latombe URL: http://robotics.stanford.edu/users/latombe/bio.html 3. Institution: Computer Science Robotics Laboratory, Stanford University URL: http://www.stanford.edu/stanford.html 4. Research Area: On-line Robot Motion Planning 5. Summary: I'm interested in robot motion planning in general. My current research emphasize on how to bring robot motion planning on-line for dynamic environments. More specifically, I consider the scenario of a compact robotic workcell equipped with two SCARA-type manipulator arms fetching objects from a conveyer belt. The geometry of all the objects in the workspace is known in advance but everything else can be changed on-line. The problem is challenging since planning can only take very small amount of time before objects leave the workspace, and the planning time needs to be accounted for during the planning process. We approach this on-line multi-arm manipulation planning problem by decomposing the problem into four subproblems for which we developed very fast planning primitives. We have implemented our algorithm in software that simulates the robot motion and tests the on-line performance of our planner. 1. Name: Mark K. Long: long@robby.caltech.edu, long@telerobotics.jpl.nasa.gov 2. Supervisor: Joel W. Burdick: jwb@robby.caltech.edu 3. Institution: California Institute of Technology 4. Research Area: Locomotion, Sensor Based Distributed Control, Central Pattern Generators, Complex Systems...... 5. Summary: I Former Work: Kinematics and Control of Redundant Manipulators, Local/Remote Supervised Autonomy for systems with Time-Delay As member of the Technical Staff at NASA/JPL for 5 years I worked on the Kinematics and Control of Redundant Manipulators, developing Approaches for the control of the Robotics Research Arm with Composite Jacobian and Damped Least Squares Techniques. I also worked in the Supervisory Telerobotics Lab on combining Impedance Control, Generalized Compliant Motion, and Redundancy Resolution. This work included a control system for supervised autonomy with time delay, and some virtual sensing as well as distributed monitoring. II Current Work: Algorithms for Locomotion based on Central Pattern Generators and Distributed Sensor Based Control. (beginning 1993) The leg motion patterns of many 4,6,... legged animals have been shown to correspond to the stable limit cycles of coupled non-linear oscillators. Where are currently examinging this behavior as well as trying to understand robustness issues, changes in the oscillation pattern during turning motion, and the role of sensor feedback in the success of the control. Additionally, some aspects of complexity theory arise when examing the emergent behavior of the entire system of simple local controllers for each leg. It is resonable to ask: how does one design simple sensor based local controllers for each leg that when combined with the other legs through a simple central pattern generator has the emergent bahvior of stable walking and turning at a variety of speeds. 1. Name: Douglas C. MacKenzie email: doug@cc.gatech.edu Mosaic URL "file://ftp.cc.gatech.edu/pub/ai/students/doug/Doug.MacKenzie.html" 2. Supervisor: Ronald C. Arkin email: arkin@cc.gatech.edu 3. Institution: Georgia Institute of Technology, Atlanta, Georgia, USA 4. Research Area: Behavioral planning, mobile manipulation. 5. Summary: Behavior-based robot architectures are systems where the overt behavior of the system emerges from the complex interactions of numerous simple sensorimotor behaviors. The distributed nature of the overt behavior generation enormously complicates the problem of configuring the system to generate a desired overt behavior. Instead of modifying a single object, a set of sensorimotor behaviors must be selected and parameterized (a configuration) such that an appropriate overt behavior is manifested. This research will automate the process of generating a behavior configuration by creating an interactive, graphically based, configuration designer. The designer will function as an assistant, capable of pointing out areas of the design intentions which are not met by the current configuration, suggesting additions, deletions, and modifications, as well as insuring syntactic validity, semantic validity, and sufficiency of the final design. Configurations will be represented in the Configuration Description Language (CDL), a context free language which has been developed to allow compact, exact description of individual robot configurations as well as the interactions of societies of cooperating mobile robots. An optimizer module will verify that each member of the generated configuration is necessary, and also that the resulting configuration is sufficient with respect to the designer's intentions. Architecture specific code generator modules will allow generating C code using various methodologies (i.e. Subsumption, Schemas, etc.). Name: Amol Dattatraya Mali Supervisor: Amitabha Mukerjee. ( amit@iitk.ernet.in ) Institute: Centre For Robotics, Indian Institute of Technology, Kanpur, Uttar Pradesh, India, 208016. SUMMARY: We have identified stimulus overgeneralization as the cause of cyclic behaviour. We have analyzed cyclic conflict in this research. We have adopted a 3-tuple model of behaviour in which we express a behaviour module by < s, a, c> where s is stimulus, a denotes action and c denotes the consequence. We have developed a notation for power, usefulness, flexibility of behaviours. In practice to do tasks behaviours need to be triggered in a particular sequence where stimulus of each behavioural module in the chain is implied by the consequence of the module immediately preceding it. A cyclic conflict occurs when the consequence of a module later in the temporal chain of behaviours triggers its stimulus or stimulus of some module before it in the chain and the cycle is not terminated by suppression by higher level modules or by a termination condition. The cycle detection strategy that we have used is of forming a temporal graph of behaviours based on action sequences and performing graph search. The solutions that we have proposed to eliminate the cyclic conflict are (1) specialize the stimulus of module earlier in the chain so that the consequence of module later in the chain does not trigger it. (stimulus specialization) (2) Modify the action of module later in the chain so that its consequence does not trigger the stimulus of module earlier in the chain. (response generalization). 1. Name: Marinus Maris email: maris@ifi.unizh.ch URL: http://josef.ifi.unizh.ch/groups/ailab/people/maris.html 2. Supervisor: Rolf Pfeifer email: pfeifer@ifi.unizh.ch 3. Institution: Dept. of Computer Science University of Zurich Winterthurerstrasse 190 CH - 8057 Zurich, Switzerland 4. Research Area: Autonomous Robots, Path Planning 5. Summary: To understand and design a real autonomous robot several aspects need to be investigated. The main research strategy focusses on adaptive control structures to enable the robot to manipulate its maneuvering around in the environment. As an example we have designed a robot that avoids obstacles utilizing just one sensor. 1. Name: Fred G. Martin email: fredm@media.mit.edu 2. Supervisor: Edith Ackermann email: edith@media.mit.edu 3. Institution: Media Laboratory, Mass. Inst. of Technology 4. Research Area: Robotics in Education 5. Summary: My work is concerned with the possibility of revitalizing the modern undergraduate engineering curriculum by including intensive design workshops based on the task of creating mobile autonomous robots. Included in this work is the design of hardware and software to support such activities, and the development and analysis of appropriate classroom/workshop environments. 1. Name: George Mobus email - mobus@ponder.csci.unt.edu 2. Supervisor: Paul Fisher 3. Institution: University of North Texas, Dept. of Computer Science 4. Research Area: Synthetic Brain - Artificial Neural Network Controller 5. Summary: A new learning mechanism called the Adaptrode, more closely emulates real biological synapses in encoding both activity- dependant and associative information with causal temporal ordering of conditionable to unconditionable stimuli. Embedded in an artificial neuron, and that in a network, this learning mechanism has been shown to mimic much of the classical conditioning paradigm [Mob94]. Such networks form the basis of an artificial brain in a Braitenberg vehicle [Bra84], #11, that learns to associate various non-semantic environmental cues to sensory stimuli that have meaning in the sense of reward or punishment [MF94]. The robot can also learn short-term associations which are contrary to long-term conditions without interfering with or washing out the latter [Mob94]. [Bra84] Valentino Baitenberg, "Vehicles: Experiments in Synthetic Psychology," The MIT Press, Cambridge, 1984. [Mob94] George Mobus, "Toward a theory of learning and representing causal inferences in neural networks," in D.S.Levine & M.Aparicio (Eds.) "Neural Networks for Knowledge Representation and Inference," Lawrence Erlbaum Assoc., Hillsdale NJ., 1994. [MF94] George Mobus & Paul Fisher, "MAVRIC's Brain," To be presented at Industrial & Engineering Applications of Artificial Intelligence & Expert Systems Conf., May 31 - Jun 3, 1994, Austin, TX. Proceedings to be published by Gordon and Breach Science Publishers. 1. Name: Simon P. Monckton email:monckton@mech.ubc.ca 2. Supervisor: D. Cherchas email: cherchas@cs.ualberta.ca 3. Institution: University of British Columbia, B.C., Canada. 4. Research Area: Multiagent Robotics 5. Summary: Most industrial manipulators employ a mapping between joint space and cartesian space either in the form of an inverse kinematic solution or the Jacobian inverse. This approach has evolved out of the understanding of kinematics and dynamics of mechanisms and now is the exclusive manipulator control methodology. Unfortunately, these approaches require significant support by world and dynamic models to achieve robust performance under varying environmental conditions. Furthermore, redundant manipulation often makes these approaches impractical to the point where few manufacturers consider the development of manipulators with greater than 6 d.o.f.. This research addresses a new possibility, a cooperative architecture of intelligent agents contributing toward the pursuit of a global objective while pursuing local objectives. A literature survey and early simulations indicate that this approach not only viable, but less compute intensive than existing adaptive and redundant control methods. 1. Name Jane Mulligan (mulligan@cs.ubc.ca) 2. Supervisor Alan Mackworth (mack@cs.ubc.ca) 3. Institution University of British Columbia, B.C., Canada 4. Research Area Integration of Sensing and Action 5. Summary My work looks at the sensory and model information required to achieve robotic tasks and proposes a layered structure for integrating sensing and action. Layers are organized based on the increasing informational/environmental complexity of 5 basic classes of tasks. NAME : Elizabeth Nitz enitz@mines.colorado.edu SUPERVISOR : Dr. Robin Murphy rmurphy@mines.colorado.edu INSTITUTION : Colorado School of Mines, Golden, Colorado RESEARCH AREA : Multiple Mobile Robotics & Communication SUMMARY : Current work focuses on constructing a new collaborative robot architecture consisting of one computationally-powerful "master" robot, who learns all it needs to know about a particular environment and possibly produces plans, and multiple less-powerful "apprentice" robots who then use the knowledge gathered by the master to carry out the plans or tasks within the environment. This type of system should combine the robustness of multiple homogeneous robots via redundancy with the cost-efficiency (and possibly disposability) of simple robots, and at the same time use the latest in computationally-intensive perceptual algorithms for learning. The target application is exploration and monitoring of hazardous or unfriendly environments. 1.- Name: Vicente Parra-Vega email:vega@arimotolab.t.u-tokyo.ac.jp 2.- Supervisor: Suguro Arimoto email: 3.- Institution: University of Tokyo, Tokyo, Japan 4.- Research Area: Control of Robot (adaptive/VSS/discontinuous/robustness) 5.- Summary: The research has focused on controlling robot manipulator for free (position) and constrained motion (force/position) as well by means of nonlinear techniques. Adaptive control. Discontinuous adaptive control. Variable structure control. Adaptive VSC. PD. Robustness. Stability (asymptotic and exponential). It takes into account the nonlinear model of the robot manipulator plus friction forces and the dynamics of the motor at each joint. No experiments, only theoretical work and computer simulations. 1. Name: Miles Pebody. e-mail: M.Pebody@cs.ucl.ac.uk 2. Supervisors: John Campbell. e-mail: J.Campbell@cs.ucl.ac.uk John Gilby. e-mail: 100115.624@compuserve.com 3. Institution: University College London, UK. 4. Research Area: Applied intelligent sensing and control 5. Summary: My project deals with aspects of intelligent sensing and control in a system of functionally and physically distributed control elements that are embedded and situated in a real-world environment. The system is an active laser scanning sensor device used in industry for analysing and detecting defects in products such as glass, plastic film, metal and painted surfaces which are moved through its laser beam. Reflected laser light is detected by a number of different sensors and information interpreted to locate and identify defects. This in turn can be used to direct the production process to deal with any critical situation detected. The aims of the project are to develop and explore the nature and effectiveness of the techniques used in Behaviour Based Artificial Intelligence when applied to a real world environment other than that of mobile robotics. The initial aim of the work is to develop a Subsumption Architecture based control mechanism and then to expand on initial results by exploring aspects of agent cooperation and learning. New control strategies will be experimented with which aim to increase the reliability and robustness of the system. 1. Name: Giovanni Cosimo Pettinaro 2. Supervisor: Chris Malcolm 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Behaviour Based Approach in Assembly Robots 5. Summary: Investigating the existance of a set of atomic behaviour with which describing any kind of more complex behaviour. 1. Name: Chris J. Pudney email: chrisp@cs.uwa.edu.au 2. Supervisor: Prof. Robyn Owens email: robyn@cs.uwa.edu.au 3. Institution: Univ. Western Australia, Nedlands 6009, Western AUSTRALIA 4. Research Area: Surface modelling for sensor equipped robots 5. Summary: A robot equipped with range sensors moves its sensors over the surface of an object. The sensor data obtained from the sensors is used to construct a model of the surface, and the surface is used in turn to control the robot's motion. Thus the surface model is constructed on-line. 3-D surface modelling techniques and algorithms for controlling the robot's surface following motion are being developed. Name : Ranganathan Ramanathan (AKA Rungun) email : ramanath@asel.udel.edu Supervisor : Dr. Rahmim Seliktar & email : seliktr@duvm.ocs.drexel.edu Dr. Tariq Rahman email : rahman@asel.udel.edu Institution : Drexel University, MEM department, Philadelphia, PA 19014, USA Research Area : Rehabilitation Robotics Summary : Design and development of powered orthosis. Stuck at an interesting but tough problem of finding out an GOOD anti-gravity mechanism to use. Then we power this mechanism, and look into various control issues and human machine interface. 1. Name: Dan S Reznik email: reznik@robios.me.wisc.edu 2. Supervisor: Vladimir Lumelsky email: lumelsky@robios.me.wisc.edu 3. Institution: University of Wisconsin-Madison, Madison, WI, 53706, USA 4. Research Area: Sensor-based motion planning for highly-redundant kinematic structures 5. Summary: I am working on the design of sensor-based algorithms for for highly-redundant robots -- so far I have considered snake-shaped robots, multi-finger hand with lots of links per finger, and "multi-branch" snakes, which are tree-shaped robots with lots of degrees of freedom. We consider both planar and 3D structures. 1. Name: Julio Kenneth Rosenblatt email: jkr@ri.cmu.edu 2. Supervisor: Chuck Thorpe email: cet@ri.cmu.edu 3. Institution: Robotics Institute, Carnegie Mellon University Pittsburgh, PA, USA 4. Research Area: Mobile Robot Architectures 5. Summary: The Distributed Architecture for Mobile Navigation (DAMN) provides a framework for independent, distributed, task-achieving behaviors, similar in spirit to the Subsumption Architecture. One important difference between DAMN and the Subsumption Architecture is that rather than one behavior overriding another, DAMN behaviors send weighted votes to an arbiter wheich then selects the action that best satisfies several objectives concurrently. 1. Name: Peter Scheffel email: peter@concave.cs.wits.ac.za 2. Supervisor: Conrad Mueller email: conrad@concave.cs.wits.ac.za 3. Institution: University of the Witwatersrand, Johannesburg, SA 4. Research Area: Path planning for manipulators with DOF<6 5. Summary: The aim of the research is to find genrally applicable techniques to improve the performance of path planning without precomputing the configuration space. Initially an implementation of an approach that does precompute the configuration space was attempted. This was found to have very poor performance especilly on simple cases like an empty environment! The research has therefore concentrated on finding solutions to simple problems quickly. Results achieved so for obvious paths in 3 DOF for a 2D env. take under 20 seconds, on a standard 486 PC, with some paths only taking 3 seconds. Most paths can be found in under 10 minutes and memory limitations hinder paths that could take more than an hour. Optimisations have played a large part in the feasiblity of the reasearch. Improvements of the order of 5 fold are not uncommon using compulational geometry techniques to improve the geometric intersections of lines. 1. Name: Jeff Schneider email: schneider@cs.rochester.edu 2. Supervisor: Chris Brown email: brown@cs.rochester.edu 3. Institution: University of Rochester 4. Research Area: Acquisition of Robot Motor Skills 5. Summary: In open loop skills such as throwing a ball, an entire robot control sequence can be viewed as a single point in a high dimensional space. Then, the problem of improving accuracy as well as increasing range of performance is a search problem. We have implemented a throwing robot with a flexible link. We have designed an efficient way to search the space with the result that our robot learns to "whip" its flexible link at the right frequency to produce long throws. The result is particularly encouraging since the robot was not given any model of its flexible link, and no samples of the "whipping" motion were ever shown to it. Recent work considers the acquisition and improvement of closed loop skills. Highly skilled humans have the ability to perform complex motions relatively open loop (consider a hockey player that corners in a single smooth motion compared to the beginner that must concentrate on balance throughout the turn). We believe that closed loop skill acquisition can benefit from an attempt to make the skills more open loop as learning progresses. 1. Name: Armin Sulzmann email: sulzmann@imtsg1.epfl.ch 2. Supervisor: R.Clavel email: 3. Institution: Swiss Federal Institute of Technology, Lausanne, Switzerland 4. Research Area: Micro-Robotics 5. Summary: This research examines the question: Developement of a vision-based (virtuel-reality) System to guide the manipulations of microsystems, microstructurs, etc. 1. Name: Hans Tangelder email:j.w.h.tangelder@io.tudelft.nl 2. Supervisor: Joris Vergeest 3. Institution: Delft University of Technology Faculty of Industrial Design Engineering Delft, The Netherlands. 4. Research Area: Rapid Prototyping using Robot Milling 5. Summary: One of the favourable techniques for rapid shape prototyping is numerically controlled milling. For this purpose we have installed a Sculpturing Robot work cell consisting of a 6-DOF industrial robot and a turn table on which a foam stock is placed. The Sculpturing Robot work cell is driven by an off-line generated path file. This path is calculated from the data representing the shape, or object, that must be replicated. This path is the result of a strategy that must take into account the limits of tool work space as well as the shape and placement of all obstacles within this space. The proposed project aims at finding appropriate methods for such a process. They must meet severe requirements concerning safety, autonomy, robustness, accuracy, speed and insensitivity to imperfections of the input geometry. 1. Name: Eddie Tunstel email: tunstel@chama.eece.unm.edu or tunstel@robotics.jpl.nasa.gov 2. Supervisor: Dr. M Jamshidi email: jamshid@houdini.eece.unm.edu 3. Institution: University of New Mexico, Albuquerque 4. Res Area: Fuzzy and Intelligent Control of Mobile Robots 5. Summary: This research focusses on the development of hybrid intelligent control architectures for autonomous mobile robots and mobile manipulation. The work includes investigations of various combinations of paradigms such as fuzzy logic, neural networks, behavior control, and genetic algorithms for real time motion control. The research focus is on control architectures for navigation, path planning, and environment mapping with empahasis on embedded application. 1. Name: Cem Unsal email: unsal@blackbox.cl.ee.vt.edu 2. Supervisor: John S. Bay email: bayj@vtvm1.cc.vt.edu 3. Institution: Virginia Tech, Blacksburg, VA, USA 4. Research Area: Multiple Mobile Robots (Army-ant Project) 5. Summary: My research is based on the idea of using a large homogeneous population of mobile robots as a transportation system. Army-ant scenario may also be applied to space/underwater missions. We treat "army-ant swarm" as a self-organizing system. Robots are simple in terms of knowledge and/or communication abilities. Important characteristic of the scenario are: lack of map knowledge, large number of agent, non-hiererchical structure, emergence (of some form) of intelligence from local interactions and simple behavioral rules. I'm currently working on behavioral self-organization (decision systems) of multiple agents. 1. Name: Jim Vaughan email: jev1@uk.ac.bton.unix 2. Supervisor: Graeme Awcock email: gja@unix.bton.ac.uk 3. Institution: University of Brighton, UK 4. Research Area: Machine vision/sensor fusion 5. Summary: Using intelligent sensors, particularly vision, to generate an accurate perception of a machine's environment, to allow it to operate with a high degree of reliability 1. Name: Virgilio B. Velasco Jr., aka "Dean" email: vbv@pris.eeap.cwru.edu 2. Supervisor: Wyatt S. Newman email: wsn@pris.eeap.cwru.edu 3. Institution: Electrical Engineering and Applied Physics Department Case Western Reserve University AND Center for Automation and Intelligent Systems Research 4. Research Area: Agile Manipulation 5. Summary: Agile manufacturing is a revolutionary product assembly approach which allows the manufacturing of a wide variety of products, with selectable features and batch sizes. For a robot to do highly agile manufacturing, it must also be agile in its ability to grasp a wide variety of objects. This research project will focus on techniques for grasping and manipulating many different objects with a minimal selection of grippers and external fixtures. It will also seek to grasp such objects within a minimal amount of time. 1. Name: Richard Voyles email: robodude@cmu.edu URL: http://www.cs.cmu.edu:8001/afs/cs.cmu.edu/user/deadslug/ftp/home.html 2. Supervisor: Pradeep Khosla email: pkk@ri.cmu.edu 3. Institution: Carnegie Mellon University, Pittsburgh, PA, USA URL: http://www.cs.cmu.edu:8001/afs/cs.cmu.edu/user/mwgertz/www/aml.html 4. Research Area: Multi-Agent Control/Perception 5. Summary: I'm investigating the cooperation of relatively dumb agents with minimal communication channels during control and perception tasks. I'm applying systems of encapsulated agents to control of a Utah/MIT dextrous hand, control of a Puma 560, and possbily to the task of selecting control methodologies for a robot. 1. Name: Gabriel D. Warshaw email: gabriel@sce.carleton.ca 2. Supervisor: Howard Schwartz 3. Institution: Carleton University, Ottawa, Ontario, Canada 4. Research Area: Sampled-Data Robot Adaptive Control 5. Summary: I am addressing the stability and performance of discretized adaptive control algorithms for robotic manipulator control, and the compensation of these algorithms for improved stability and tracking performance. The discretization of adaptive control algorithms published in the literature can result in a sampled-data robot system for which stability has not been guaranteed. By formulating the entire sampled-data system in continuous-time, I have used Lyapunov's direct method to determine the stability and to derive a non-linear discrete-time compensating term. I have demonstrated the theoretical results through simulation and implementation on a 2 degree-of-freedom direct drive manipulator. 1. Name: Martin D. Westhead 2. Supervisor: John Hallam 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Theoretical underpinnings of behaviour based systems 5. Summary: This work is still in its early stages, but will attempt to apply formalisms such as process algebra's, dynamical systems and petri nets to the problem of behaviour based robotic control. The goal of the work is a better understanding of the parallel interaction of behaviour based systems with the hope that this might provide the foundation for a more rigorous design methodology than those currently employed. 1. Name: Jeremy Wyatt 2. Supervisor: Gillian Hayes and John Hallam 3. Institution: Dept of Artificial Intelligence, Edinburgh University, UK 4. Research Area: Learning in Mobile Robots 5. Summary: I an interested in using learning to help design robot controllers. I am working on combining a technique called reinforcement learning with a behaviour based controller. Elementary behaviours are learned separately, and then coordinating behaviours are learned. Ultimately the aim is to build a hierarchy of behaviours with several levels. 1. Name: Gordon Wyeth email: wyeth@elec.uq.oz.au 2. Supervisor: Mark Schulz email: marks@elec.uq.oz.au 3. Institution: Robotics Laboratory, Dept. of Electrical and Computer Engineering, University of Queensland, Brisbane, Australia. 4. Research Areas: Artificial Neural Control of a Hunt and Gather Robot, Micromice. 5. Summary: Hunt and Gather Robot The main aim of this project is to produce models of cognitive structures that support intelligent behaviour sufficient to allow a mobile robot to perform collecting tasks. The project involves the construction of a robot dog, CORGI, that retrieves tennis balls around the lab. The techniques used to build this robot are similar to those used in behaviour-based robotics, but are based on a neural paradigm. The robot's primary sense is a CCD camera that is used to locate the tennis balls, sense obstacles and to provide visual cues to allow the robot to return to its home position. The artificial neural network architecture is based on a combination of conventional networks (MLP, SOM) and constructs found in Braitenberg vehicles. Micromice Micromice are autonomous maze solving robots that are entered in competitions to see who can solve the maze and run the fastest path in the least time. My Micromice include CUQEE I & II, Australian Micromouse Champions. CUQEE III is currently in development and should debut at the end of 1994. 1. Name: Brian Yamauchi email: yamauchi@alpha.ces.cwru.edu url: ftp://alpha.ces.cwru.edu/pub/agents/yamauchi/yamauchi.html 2. Supervisor: Randall Beer email: beer@alpha.ces.cwru.edu 3. Institution: Department of Computer Engineering and Science Case Western Reserve University, Cleveland, OH 4. Research Area: Exploration and Spatial Learning in Dynamic Environments 5. Summary: The goal of my Ph.D. thesis research is to develop techniques that will allow mobile robots to explore, learn, and navigate in the presence of unpredictable moving obstacles and dynamic changes in both the topology and structure of the environment. I have developed an exploration and navigation system that combines reactive behaviors for low-level control with an adaptive place network for spatial learning. This network consists of place units that use competitive learning for place recognition along with adaptive connections that learn the geometric relationships between these places. This system has been implemented on a real Nomad 200 mobile robot equipped with sonar, infrared, and laser range sensors, at the Navy Center for Artificial Intelligence at the Naval Research Laboratory. 1. Name: Mark Yim email: mark@killdeer.stanford.edu 2. Supervisor: J.C. Latombe email: latombe@cs.stanford.edu 3. Institution: Stanford University, Stanford CA, 94305 4. Research Area: Reconfigurable Modular Robot Locomotion 5. Summary: A dynamically reconfigurable modular robot named Polypod has been designed, simulated and partially constructed. Research is being done on unusual statically stable locomotion gaits implemented on Polypod, for example, a rolling loop, a moving carpet with many feet, slinky locomotion... Each gait is achieved with a very simple behaviour based control scheme. A taxonomy of locomotion and the kinematics of locomotion will be analyzed.