M.-R. Akbarzadeh-T., M. Jamshidi, M. Johnson, and N. Vadiee
NASA ACE Center, EECE Building
University of New Mexico
Albuquerque, New Mexico 

Robotic issues in waste handling are addressed by considering both research and educational missions of Waste-management Education and Research Consortium (WERC). During the research aspect of this project, several robotic test-beds are developed which facilitate experimental research in the area of robot control, and in particular intelligent robot control based on soft-computing approaches with application to hazardous waste management. In addition, various issues in robot control such as flexible links, trajectory following in unstructured environments, and autonomous mobility are investigated by applying soft computing paradigms such as fuzzy logic, neural networks, and genetic algorithms to the test-beds. These controllers are capable of learning and adaptation and therefore can improve robot performance in cluttered and unstructured environments such as in waste handling applications. Soft-computing approaches provide a new generation of smart robots which incorporate human expert knowledge with learning and adaptation capabilities for optimal performance. Fuzzy control is specifically applied to highly nonlinear systems which lack a simple analytic model. During the educational aspect of this project, the experimental test-beds are used to train students from Southwestern Indian Polytechnic Institute (SIPI) in advanced intelligent and autonomous robot control techniques. Students from SIPI also participate in various aspects of development and evaluation of the test-beds. As a result of this project, a formal course curriculum in robotics is integrated with SIPI's existing technology program. Finally, students from Southwestern Indian Polytechnic Institute (SIPI) participate in mentorship and summer internship programs coordinated among SIPI and Center for Autonomous Control Engineering.  

Robots are man made machines for the purpose of relieving human from performing undesirable, repetitive and/or difficult physical tasks. Handling hazardous waste is such a task which, in addition to its inherent hazards to human life, is unstructured and requires increasingly higher precision and reliability. As a result, robotics has many applications in the field of waste management. As a result of the increasing demand for increased robot performance in unstructured environment of a waste site, several interesting issues deserve to be investigated. These issues include control of flexible robots, telerobotics, robot navigation in unstructured and unmapped environments, and redundant robots. For example, the Department of Energy's underground storage tanks for hazardous waste material are 80 feet in diameter and 35 feet in depth while the access hole where a robot can enter is only 12 inches. Consequently, the robot's cross sectional radius must be less than 12 inches, yet its arm should be able to span the large tank. This requires a very long and slender robot arm where flexibility plays a vital role in robot dynamics. In a situation as above, neglecting flexibility can result in instability and undesired oscillations. 

Because of the unstructured nature of the hazardous waste applications, a reliable and accurate mathematical model is not always available. As a result, conventional methods of control may not be the most appropriate approach since such methods tend to heavily depend on accuracy of the mathematical representation. Fuzzy logic is a method of approximate reasoning. It allows human expert knowledge to be "downloaded" to the machine's processor, thereby allowing the machine to process information similar to how a human operator normally would. Machines equipped with a fuzzy logic engine can therefore benefit from either or both conventional model based techniques and/or human operator's intuition and experience. This translates to a more robust and "intelligent" robotic system capable of performing complicated tasks in unstructured environments with noisy and/or unknown parameters. 

Fuzzy logic is one of several soft-computing paradigms. Other soft-computing paradigms such as Neural Networks (NN) and Genetic Algorithms (GA) also contribute to the added machine intelligence in different ways. Neural networks are used in various application areas such as in pattern recognition and nonlinear model learning. Genetic algorithms are optimization algorithms modeled after Darwinian model of nature's evolutionary process. Since GA's optimize processes by function evaluation rather than gradient evaluation, they are suitable for optimizing knowledge base of fuzzy controllers. A fuzzy knowledge base is comprised of several parameters such as membership functions and a set of rules which can be optimized using GA's. 

In this project, robotic issues in waste handling are addressed by considering both research and educational missions of WERC. During the research aspect of this project, several robotic test-beds are developed which facilitate experimental research in the area of robot control, and in particular intelligent robot control based on soft-computing approaches with application to hazardous waste management. In addition, various issues in robot control such as flexible links are investigated by applying soft computing paradigms such as fuzzy logic, neural networks, and genetic algorithms to the test-beds. These controllers are capable of learning and adaptation and therefore can improve robot performance in cluttered and unstructured environments such as in waste handling applications. Soft-computing approaches provide a new generation of smart robots which incorporate human expert knowledge with learning and adaptation capabilities for optimal performance. Fuzzy control is specifically applied to highly nonlinear systems which lack a simple analytic model.  

During the educational aspect of this project, the experimental test-beds are used to train students from Southwestern Indian Polytechnic Institute (SIPI) in advanced intelligent and autonomous robot control techniques. Students from SIPI will also participate in all aspects of development and evaluation of the test-beds. As a result of this project, a formal course curriculum in robotics is integrated with SIPI's existing technology program. Finally, students from Southwestern Indian Polytechnic Institute (SIPI) will participate in mentorship and summer internship programs coordinated among SIPI, NASA ACE and the team of WERC researchers.

The test-beds are a unique facility which could be utilized for future research projects in intelligent control technology. The test-beds will also be utilized to educate students in intelligent and autonomous robot control. The test-beds consist of a single link flexible arm, a four axis ADEPT II robot, and a mobile robot donated by JPL. Real time systems based on a high speed Texas Instrument's TMS320C30 Digital Signal Processor board and a 486DX2 PC 66 MHz is developed to experimentally evaluate the performance of these controllers. This paper will also report on development of a new software for simulation and experimental research in soft computing technologies, SoftLab. Softlab is written in Visual C++ and provides a graphical user friendly environment.

This paper reports on several different active research initiatives as is detailed in [1-13]. Here we briefly introduce those areas as follows. First, the flexible robot test-bed is discussed in the following section. Then, the various hybrid intelligent methodologies are illustrated. SoftLab is discussed next. SoftLab is a soft-computing environment which aims to provide a fast and friendly interface for development of new intelligent control architectures. Finally, the course taught at SIPI is reported. 

Fuzzy logic is strongly associated with experienced and practical knowledge in handling imprecise systems operating in unstructured environment of the real world. As result, developing test beds and obtaining experimental data comprises an important aspects of developing fuzzy logic systems.

In this section, we will discuss various aspects of the experimental set up such as the electronic hardware, the mechanical arm. The software algorithm will be discussed in the next section.  

The mechanical design for the flexible robot includes the flexible link, 1 meter long x 10 cm high x 1/16 of inches wide. A rigid rod is used in the tip sensing apparatus between the center of the motor shaft and the tip. The rod is assumed to be rigid because it does not carry any load is made of Carbon composite material. The assumption of lightness is useful if one decides to neglect the loading effect of the rod in system modeling. The position encoder for the tip is bolted to the motor mounting structure. It has a hollow shaft which allows the motor shaft to go through. 

To accommodate automatic control algorithms, e.g. the fuzzy logic control, strain gages are selected to provide a measure of strain at different points of the link. The required components for the flexible link tested are listed Table I.

DSP Research's TIGER 30 board utilizes Texas Instruments TMS320C30 DSP and is fully compatible

with a Pentium-133 Processor board. The TMS320C30 digital signal processor has a powerful instruction set, operates at 40MFLOPS, and can process data in parallel. The TIGER board has the ability of being interfaced with a PC therefore allowing dual processing of the genetic algorithm as well as the fuzzy logic controller. The fuzzy logic controller is continuously processed in the DSP board at 1 kHz. Parameters which identify the rule-base (membership functions and fuzzy Associative memory) are stored in a block of memory residing on the DSP board. This block of memory is modified by the PC which runs the code for genetic algorithms. Each time the PC has completed an optimization task, a signal is sent to the DSP board. The DSP board then modifies its parameters accordingly and send data on the performance of the fuzzy controller back to the PC. A generic data acquisition system was designed to serve as a medium for data transfer between the control system and the DSP board. The data acquisition system is capable of asynchronous and 16-bit data transfer. A program written in C is used to transfer data to and from the control system.

softball is a software environment for research and development in intelligent modeling/control using soft-computing paradigms such as fuzzy logic, neural networks, genetic algorithms, and genetic programs. SoftLab addresses the inadequacies of the existing soft-computing software by supporting comprehensive multidisciplinary functionalities from management tools to engineering systems. Furthermore, the built-in features help the user process/analyze information more efficiently by a friendly yet powerful interface, and will allow the user to specify user-specific processing modules, hence adding to the standard configuration of the software environment. 

Computers and computer software play an increasingly significant role in our modern world. This is particularly evident in the process of research and innovation where complex systems are analyzed and controlled. Intelligent control/modeling is one of the research areas which has received a great deal of attention and interest among researchers in science and engineering in recent years due to its ability to incorporate human knowledge and reasoning ability in computing processes. The fundamental basis of such approaches are expert systems, fuzzy logic, neural networks, and evolutionary computations, collectively named the "soft computing" paradigms. Soft computing paradigms and their hybrid combinations with other approaches are commonly used to enhance artificial intelligence (AI) and incorporate human expert knowledge in computing processes such as in the design of intelligent autonomous systems/controllers and handling complex systems with unknown parameters such as in prediction of world economy, industrial process control, prediction of geological changes within earth ecosystem, and robot control in space for scientific research or on earth for handling hazardous waste. 

Even though soft computing paradigms can potentially provide answers to a wide array of today's problems, there is not yet a comprehensive software environment which can address the multidisciplinary functionalities expected from soft computing paradigms. Currently, there are several software environments in the areas of fuzzy logic, neural networks, and evolutionary algorithms. These include Togai Infra Logic [14], Fuzzy Inference Design Engine (FIDE) [15], Fuzzy Logic Development Kit (FULDEK) [16], and O'INCA [17]. These software environments, however, are each developed for only a limited capability and are application specific. None of the current software allow users to improve the functionality of the software by incorporating new user-defined modules. They do not utilize all of the soft-computing paradigms in a single software environment. And finally, none of them support add-on hardware modules/processors for real-time experimentation and improving computational efficiency. As a result, researchers, in the area of intelligent systems, often devote a significant amount of resources to develop their own application specific software and data acquisition hardware, making it time-consuming and costly to perform research and development in this area. 

This section addresses the inadequacies of the current software by proposing a comprehensive soft-computing based software environment which can address multidisciplinary functionalities from management tools to engineering systems. The software environment, SoftLab, will be written using Visual C++ for a user friendly environment. It will be a general purpose development/analysis environment where the inadequacies of the existing soft-computing software will be addressed. Furthermore, the built-in flexibility in SoftLab will help the user process/analyze information more efficiently by a friendly yet powerful interface, and will allow the user to specify user-specific processing modules, hence adding to the standard configuration of SoftLab. It will also allow applications in a wide range of directions thereby easily adding artificial intelligence to areas where AI has not yet been explored.  

SoftLab 1.b is the first preliminary release which simulates fuzzy logic control systems, as depicted in figure 3, of the following properties:

• The system:

1. The system can be either linear or nonlinear and of any desirable dimension.
2. The system equations can only contain mathematical functions defined in Visual C++ and must be presented in state space form.

• The controller:

1. Controller should be a Fuzzy Logic Controller (FLC).
2. FLC can have any number of rules and membership functions.
3. Membership functions are triangular shaped.
4. Antecedents are joined using AND and/or OR.
5. Min/Max is used for rule evaluation.
6. Center of Gravity is used for defuzzification of rules. 

Figure 4 illustrates a snapshot of the current version of SoftLab.

The following features will be added to the future versions of SoftLab, figure 6.

• Integrate fuzzy logic, neural network and evolutionary algorithms in one development environment where adaptation and machine learning can be performed.
• Develop a modular system where a third party can integrate an application specific module to the software environment. As a result, other methods of artificial intelligence can be incorporated in the future.
• Capability to add hardware modules such as Digital Signal Processors (DSP), parallel processors, data acquisition hardware, and remote processors through the Internet. This will address two essential purposes:
• Addresses computational efficiency through multi- and parallel processing environment.
• Allows real-time experimentation and data gathering.
• Interface to knowledge based software and multi-media tools. The strength of a given software is increased significantly if it can take advantage of the computational power of other existing software. One of our objectives is to add soft-computing techniques to data-base management tools such as Oracle.
• Application areas such as in control systems, data-base management, dynamic simulations, and real-time experiments.  

In summary, SoftLab is intended to respond to a growing demand for a user friendly, dynamic and powerful software environment for research in intelligent control. In this paper, various current and future aspects and features of SoftLab are discussed. SoftLab not only provides a good simulation environment, but also has several features which facilitates experimental research. At this time, only a limited test version is released for the purpose of testing and debugging. The next version of SoftLab is expected to be ready for release to the public.  

An introductory course in robotics is taught to students at the Southwest Indian Polytechnic Institute in an effort to: introduce them to robot technology, interest them in the sciences, and interest the students to consider pursuing a course of study at the University of New Mexico in robotics and/or related sciences. Topics covered include robot history and evolution; types; components (links, drives, sensors, peripherals); component integration; applications in industry, science, and space and hazardous environments; and Artificial Intelligence and its use in robotics. In addition, student projects are performed in which the students build a simple robot, make it work, get it to do something, write about the experience, and present their experience and work to the rest of the class. 

A basic overview course on robotics is taught at the Southwest Indian Polytechnic Institute (SIPI), Albuquerque, New Mexico during the Fall 1997 term. The objectives of the course are to 1) introduce robot technology, 2) provide them hands-on experience with robots, 3) interest students in pursuing further studies at a four-year university in a science related curriculum, 4) give the students an idea of the robotics related work being performed at the University of New Mexico, 5) and provide them insight into how robots are used in hazardous environments and some of the problems involved. Each of these objectives are met with a high level of interest for both continuing the class and providing an advanced course. In addition, several students have expressed interest in pursuing studies in science at the University of New Mexico and other four-year universities. The rest of this article provides a summary of the topics taught along with student and instructor experiences. 

The course consists of both lecture and laboratory sessions. Lectures begin by covering robotics history starting with Egyptian Clepsydras, or water clocks in 1500 B.C., and the steam driven Hercules throwing a spear at a dragon by Hero of Alexandria in 300 B.C. to Regiomontus's iron fly (which would fly around the room and land on his hand) and Eagle which flew before Charlemagne into Nuremberg. Modern day robots including one of the first know devices to use feedback, the gyrocompass, to the various robots developed for planetary exploration are covered. In addition, some robot philosophy from Asimov's three laws of robotics to modern cognitive robot control paradigms are covered. Of course, the fact that the word robot originated in Karel Kapek's play, "Rossums' Universal Robots" and not a George Lucas film is also presented. 

Lectures continue to cover robot anatomy, structures, drive mechanisms, sensors and control methodologies. Advantages and disadvantages, capabilities and limitations of structural configurations, drives, sensors, and control methods are covered. The various artificial intelligence (AI) tools are covered with emphasis on their application to making robots more usable in uncertain or hazardous environments. Emphasis is placed upon covering fuzzy logic and artificial neural networks and their usage in adaptation of robots to unknown environments. In addition, the usage of various robot types for various hazardous environments is covered with examples from nuclear waste management (e.g. Three Mile Island). 

Laboratory work consists of five student teams building robots from kits and getting them to perform some task(s). This valuable experience enables them to see what it take to make a robot and how the different components function to make the entire system work. With most kits, some components were missing; which, provided even more experience in realizing when they reached a point at which they needed either more information, tools, or hardware to continue the project. Each student wrote a report in which they document their experiences during the project. Each team presents their project to the rest of the class and identifies what problems they encountered and what they did to overcome them, as well as, what tasks they had their robot perform and the outcomes of those attempted tasks. One item each team presented was how they would apply their robot to a hazardous environmental problem. 

One group which built a three degree of freedom, vertically articulated arm proposed using theirs in taking apart chemical munitions for hazardous chemical disposal operations. Another team with a mobile robot, proposed putting a camera and other sensors on it and sending it into an hazardous and/or nuclear situation to see what was going on so that humans would not have to go into the dangerous environment. Another group which built a hexapod decided theirs could go places the other mobile robot could not. The other mobile robot had wheels and theirs had six legs and could; therefore, climb stairs or other surfaces the other could not. They proposed perhaps having both of them around to go into a hazardous situation to cover all the potential means of getting around to see what was going on in the area. 

The student project reports and presentations represent 50 percent of their grade with two exams providing the other 50 percent. In addition to teaching students about robotics, tutoring sessions are held prior to class to teach those without prior computer experience about computers, how they work, and how to program them in BASIC. The robots built during the projects are directed using BASIC language instructions, which gives the students direct feedback on what they program their robot to perform. Overall, this course is very valuable in achieving each of the stated course objectives. 

Feedback from students was very positive. Most felt they had learned more about how things go together than from any other experience. Several commented on how much went into making a simple robot and getting it to do something. One student, an employee at Intel Corporation said she now felt qualified to go for advancement to technician grade. Several students have given their resumes and we hope to be able to bring them to the University of New Mexico to perform research. The experience was both a learning experience and a rewarding one. Delving into robot history was fun and the active class participation and lively discussions were rewarding. Each student showed pride of what they achieved in applying what they had learned in class to making their robots and getting them to operate. 

We expect to develop new intelligent robot control schemes based on new soft-computing approaches. These controllers are capable of learning and adaptation and equip robotic waste management with reliable performance in cluttered and unstructured environments. Soft-computing approaches provide a new generation of smart robotic manipulators and mobile robots which incorporate human expert knowledge with learning and adaptation capabilities for optimal performance. Fuzzy control design is specifically applied to highly nonlinear systems which lack a simple analytic model. We plan to utilize our experimental testbeds setups to train community college and university technology and engineering students in advanced robotic control techniques. Students from SIPI will participate in all aspects of development and evaluation of the experimental testbeds. These are students from SIPI's Advanced Technological Education as well as SIPI's Environmental Science and Industrial Hygiene (ESIH) Program. Dr. Nader Vadiee , a part-time educational program consultant at SIPI, have collaborated with Dr. Tom Corbitt, SIPI's chair of EISH program, in this matter.  

There are several outcomes of this research which have the potential to be commercialized. The first product is an intelligent software environment for research and development in the field of intelligent control/modeling with an special emphasis on waste management and robotics. The second items involves the introductory robotic course designed to define and address the current state of robotics in general and in waste management in particular. Solutions to robotic issues such as control of flexible robotics can also be commercialized. MediaTronix, a national minority-owned company, in close collaboration with University of New Mexico's Science and Technology Corporation (STC), will be taking the task of commercializing such potential products. 

This project involves two major aspects: research and teaching. During the research aspect, we seek innovative intelligent solutions to difficult robot control issues with respect to handling hazardous waste. The experimental test bed will serve as a valuable tool for current and future research in flexible robotics. The teaching aspect of this project introduces various aspects of robotics to students at Southwestern Indian Polytechnic Institute (SIPI). During the final stages of this project, the two aspects are expected to merge. Hence, there will be SIPI students who will actively participate in determining innovative solutions to robotic problems in hazardous waste handling. This is expected to increase students' awareness with respect to value of robotic research and also entice them to pursue higher education at UNM. 

Various basic ingredients are coming together to produce appreciable results. These ingredients include the flexible robot testbed, the intelligent control software, the intelligent control architecture, and design of a robotic course suitable for SIPI students. The flexible robot test bed is completed with its computer and DSP interface and appropriate drivers. The test bed is capable of processing to algorithms in parallel. This is particularly useful when using adaptive/optimizing algorithms for fuzzy controllers. The parallel processing allows the adaptation algorithm to be evaluated with disrupting the real-time control of the physical system. Software to run intelligent control algorithms have been developed and tested. The software allows combinations of fuzzy logic, neural networks and genetic algorithms to be incorporated in the real-time fuzzy control process. Furthermore, a hierarchical fuzzy control architecture is designed and simulated for the flexible robot. The graphical interface and hardware routines are currently being developed using Visual C++. Also, the real-time control algorithm is being developed for experimental testing.  

The introductory robotic course is successfully coordinated between SIPI and UNM. Student participation level and interest have been extremely encouraging. A solid foundation and familiarity of robotics is developed throughout the class. Moreover, students are currently involved in various hands on robotic experiments which include building a mobile robot, a multi-axis robot, and a walking robot. We expect to merge the two aspects of research and teaching next semester when students in the advanced robotic class will actively participate in research and development of robotic systems. A meeting tool place in December '97 among researchers, president of SIPI, and the chairman of department Electrical and Computer Engineering (UNM) to discuss the current state of the project and possibilities for future collaboration. Since the substantial budget cut of the original proposal, some aspects of the research has been narrowed down to more specific areas, in particular flexible robotics. With this focused area of research, we expect to make a significant contribution by completing the research which has been started in the first year, and continuing with the task of combining education and research for minority students at SIPI. 

[1] Vadiee, N., "Fuzzy Rule-based Expert Systems I and II," in M. Jamshidi, N. Vadiee, and T. J. Ross (eds.), Fuzzy Logic and Control : Software and Hardware Applications, Prentice Hall, Englewood Cliffs, NJ, 1993, Chapters 4 and 5.

[2] Vadiee, N. "On a programmable Fuzzy Logic Array (PFLA) for Soft Fuzzy reasoning Paradigms," Ph.D. Dissertation, 1995, The University of New Mexico.

[3] Vadiee, N. and Akbarzadeh, M., "Analogous Fuzzy Rule-based Expert Systems", Proceedings of FUZ-IEEE Conference, 1996, New Orleans, LA.

[4] Tunstel, E. and M. Jamshidi, "On Genetic Programming of Fuzzy Rule-based Systems for Intelligent Control," to appear in International Journal of Intelligent Automation and Soft Computing, Vol. 2, 1996.

[5] Vadiee,N., Drew J. Riedle, and M.-R. Akbarzadeh-T, " A Fuzzy Logic Autonomous Controller for the optimization of the Soil Vapor extraction Process," submitted to the WERC/HSRC joint conference on the Environment to be held in Albuquerque, New Mexico, April 22-24, 1997.

[6] M.-R. Akbarzadeh-T, E. Tunstel, and M. Jamshidi, " Genetic Algorithms and Genetic Programming: combining Strengths in one Evolutionary strategy, "submitted to the WERC/HSRC joint conference on the Environment to be held in Albuquerque, New Mexico, April 22-24, 1997.

[7] Tunstel, M.-R. Akabarzadeh-T., K. Kumbla and M. Jamshidi, "Soft Computing Paradigms for Learning Fuzzy Controllers with Applications to Robotic" North American Fuzzy Information Processing Society (NAFIPS), Berkely, CA, June 1996.

[8] M-R. Akbarzadeh-T., E. Medina, and M. Jamshidi, "DSP Implementation of Evolutionary Fuzzy Control," Proceedings of the First National Student Conference, NASA, Greensville, NC, March 1996.

[9] M-R. Akbarzadeh-T., M. Jamshidi, and P. Dorato, "Fuzzy Hierarchical Control of Distributed Parameter Systems, A Case Study on a Heating Slab," Proceedings of the Applied Computing Symposium on Applied Computing, Nashville, Tennessee, 1995.

[10] M-R. Akbarzadeh-T., K. Kumbla, S. Rodriguez, and Y. Kim "Evolutionary Fuzzy Control in Real-Time DSP Controlled Systems" Sixth International Symposium on Robotics and Manufacturing, Montpellier, France, May 1996.

[11] M-R. Akbarzadeh-T., K. Kumbla, and M. Jamshidi, "Genetic Algorithms in Learning Fuzzy Hierarchical Control of Distributed Parameter Systems," Proceedings of IEEE Conference on Systems, Man and Cybernetics, Vancouver, Canada, 1995

[12] M-R. Akbarzadeh-T. and K. Kumbla, "Intelligent Control of Desalination Plants: GA-Fuzzy Approach" First International Symposium on Intelligent Automation and Control, Montpellier, France, May 1996.

[13] Kumbla, M.-R. Akbarzadeh-T., E. Medina, "Adaptive Neuro-Fuzzy Controller a Digital Signal Processor," Sixth International Symposium on Robotics and Manufacturing, Montpellier, France, May 1996.

[14] Fuzzy C Development System, User's Manual, Togai Infra Logic Inc., 5 Vanderbilt, Irvine, CA, 92718, 1992.

[15] Fuzzy Inference Design Environment, Aptronix Inc., 2150 North First Street, Suite 300, San Jose, California 95131, 1992.

[16] Fuzzy Logic Development Kit, User's Manual, Bell Helicopters Textron Inc., P.O. Box 482, Fort Worth, Texas 76101, 1992.

[17] O'INCA Design Framework, User's Manual, Intelligent Machine Inc.

BACK