Controlled Computations

or Computer Control?
John S. Baras

Background Statements It is relatively easy to cite specific exam- open new opportunities for progress and
ples from many areas of engineering and sci- advancement in the field?
I will start this position paper with some
ence where the preceding facts manifest
statements regarding factsthat, during the It is true that our field is a relatively new
themselves in a very profound way. I will one. It has made, however, substantial con-
last 10 years, in my opinion, have dramati-
use a few just for illustration purposes:
cally changed the environment in which a tributions to engineering, science, econom-
control engineer-or, more generally, a sys- How do we control systems characterized ics,and also the standard of living inthe
tems engineer-must operate. by complex-often, poorly defined- United States and other countries. It is my
models? Typical examples include chem- opinion that the current state of the field is
The complexity of (control) systems has
ical process control, where, often,itis an excellent one. The advances in computer
increased dramatically. This is mani-
difficult to design “correct” loops and technologies provide the opportunity for vig-
fested,.for example, in tighter engineering
equations. orous and synergistic development of bal-
specifications, the need for adaptation, re-
quirements for multisensor integration, the How should one automate the operation anced new theories that combine powerful
need to account for contingencies (multi- of systems defined by precise, highly analytical tools with these technologies to
ple modalities), totally digital implemen- complex simulation models? Problems in produce beper performing systems. So, my
tations, and the need for a mix of numer- flexible manufacturing systems represent answer to the first question is unequivocally
ical and logical computations. As a result, generic examples, wherein time-prece- yes. Now, whether the research programs at
control systems design must now be dence constraints and the need for a d a p major universities and, perhaps more impor-
viewed as an interacting part of integrated tive automation further complicate design. tantly, the funding programs at various pub-
design. How should we design systems controlled lic agencies and private-sector sources are
addressing these issues is a totally different
Advances in computer technology, both in by asynchronously operating, distributed,
matter. This is where I believe some drastic
hardware [very large scale integration communicating controllers? Examples in-
changes are necessary.
(VLSI), large memories, artificial intelli- clude the computer-aided design (CAD)
Regarding the second question, the answer
gence ( A I ) chip, distributed, parallel ma- of computerkommunications networks,
dynamic capacity allocation in communi- (if you think about it hard for some time) is
chines] and software [AI, symbolic alge- relatively straightforward. Control systems,
bra, computer-aided engineering (CAE)], cation satellites, and efficient management
in general, areman-made devices that collect
have not been properly understood by of mixed traffic (voice, video, data).
data from a physical system, interpret them,
control system theorists and designers. (I How can we develop design tools for real- and produce control signals. In this process,
would even say thereis an awareness time, high-performance, non-Gaussian it is very important to pay attention not only
problem here.)Thereis, at present, a signal processors? Examples from radar, to the analytical development of the rela-
growing gap between the sophistication sonar, image, and speech signal process- tionship between control actions and observ-
and power of this technology versus its ing abound. ables but also to the models of both the phys-
use in control systems synthesis. In ad- How can one integrate multiple sensors ical system (to be controlled) as well as the
dition, the needs of control systems have for robust, digital, feedback control of man-made system (the controller). Of course,
not been properly embedded in microelec- nonlinear systems? Examples include today both can be man-made. Where I be-
tronic chip design. many-degrees-of-freedom robotic manip- lieve attention has been lacking in the last
Computers of all kinds have become in- ulators with vision, force, and pressure 10 years or so is in training, educating, and
dispensable tools, both as design aids and sensors, as well as advanced aircraft flight doing research on the empirical component
as elements of the solution (i.e., the im- controllers especially designed for the new of our science, the one that naturally ad-
plementation of a control algorithm, de- generation of unstable aircraft. dresses these very important modeling ques-
sign, or scheme). As a matter of fact, the tions. Control and systems engineering de-
design of computers and computations Two Questions pend criticaZ/y on models. This lack of
(and their formal counterparts) look more emphasis on modeling has not allowed us to
Having made these statements, let me pose
and more similar to the design of present- link early with new models of computers, or
and examine the following questions:
day control and communication systems (a VLSI devices, or even new models of com-
phenomenon that is not unexpected). Are these developments in computer tech- putation. This has led to total neglect of the
nology (both hardware and software) en- implementation question.
Presented at Santa Clara Workshop, Santa Clara hancing the future of the field of control
University,SantaClara.California,September TOsummarize this point and to answer the
system engineering, or systems engineer- second question, let me just state that, un-
18-19, 1986. Professor John S. Baras is with the
ing in general? fortunately, currently available theories and
ElectricalEngineeringDepartmentandSystems
Research Center, Universityof Maryland, College Can we identify a basic reason for the gaps design methodologies for control systems are
Park. MD 20742. mentioned and a course of action that can not in synchrony with the currently available

0272-1 70818710400-0007 $01.OO t

3 1987 IEEE
April I987 7
or planned implementation media, be it spe- needed. Similar results abound when one bility to run several modules of the design
cial-purpose chips or computers with spe- starts investigating distributed, asynchron- software system on different machines con-
cialized architectures and capabilities. More ous algorithms. As a matter of fact. I believe currently. The AI workstation provides the
precisely, the available design theories and that new mathematical, analytical, and com- direct interface with the user, the (often sym-
performance evaluation methods were de- putational advances are needed to take full bolic) problem description, and modeling.
veloped for different (now often obsolete) advantage of the new technologies. On the The multiprocessor “number cruncher” pro-
implementation media. such as analog cir- other hand. a tremendous, albeit critical. op- vides the necessary computing power for al-
cuits and sequential machines. I furthermore portunity exists for system scientists and en- most real-time execution. Finally, the graph-
claim that, although for some problems-ad- gineers to influence the chips and computers ics engine provides real-time graphics for
mittedly a small class-it is feasible to de- designed and built by the microelectronics simulation, testing, validation, and feedback
velop improved designs using the new hard- industr)... to the designer.
ware capabilities and existing theory, in the
Research Emphasis Real-7ime Implementation
majority of problems, there is a substantial
lag between the available hardware potential I will close this short position paper with This is a volatile issue, since the meaning
and its realization in the systems being built. some suggestions of where research should of “real time” relates strongly to the avail-
In my opinion, we can no longer separate be emphasized in order to capitalize the full able technologies. “Convolution” was not a
the design of a system from the implemen- development potential of our field. I will real-time operation 10 years ago, but it is
tation problem. categorize these areas by the technical (key) today. Furthermore, it is clear, at least for
Related to the issue of modeling is the words provided by the editors. control and communication systems pur-
broader problem of design integration. Em- poses, that the question is not how fast the
phasis on modeling (in the sense described Computer-Aided Engineering chip is. but rather an architectural one.
earlier) mandates emphasis on models of ac- Research is needed badly to investigate the Therefore, research in distributed, asyn-
tuators and sensors,a technology strongly effects of interactive graphics, interfaces, chronous, and synchronous architectures,
linked with computer technology. Sensor and etc., in thedesign of sophisticated CAD sys- geared toward specific control problems,
actuator technology is currently undergoing tems. Further research is necessary in order would be very beneficial.
dramatic changes,and more revolutionary to properly understand the changes needed AI Issues
developments are expected in the next dec- due to the forthcoming integration of sym-
ade. Control systems design must now be- bolic and numerical computation, which One has to be extremely careful here. I am
come a well-integrated part of the overall changes the type and form of the “knowl- not in favor of heuristics-based expert sys-
design, with strong links to other engineer- edge base” used for design. tems. However. the development and theory
ing and scientific disciplines involved; in of systems that can reason in a specified do-
System-Level Design Tools main of control engineering seems tobe
other words, design for control is now a real-
ity. What we really needis research for the rather feasible and attractive. For design sys-
development of design tools that can handle tems, AI can be very helpful in providing
integwted design from conception to imple- easy model modification, model manipula-
Computers and Control Systems tion, handling the initial intuitive and em-
mentation. Systems engineers today are
To proceed further with the analysis of the called upon to solve complex control and pirical design steps, and reducing nuisance
relationship between computers and control communication design problems for systems programming. Far more exciting are the pos-
systems. I would like to emphasize again often described by huge simulation models. sibilities offered by AI chips in the feedback
that as the years go by, they look more alike. To ask a systems engineer to solve the com- loops of control systems.
and the tremendous opportunities that open plex design problems of today without such VLSI and Architectural Issues
before us may be lost if we do not pay at- a combined arsenal of tools is similar to ask-
tention to computers and computation ing a VLSI chip designer to design the chip This is a totally implementation-related
models. As a matter of fact, this interaction without the expert CAD tools now available. topic. It is clear, however, that deeper un-
between theory, design, and implementation In this context, I have found AI languages derstanding of architectures for control and
manifests itself in many unsuspected ways. (LISP, PROLOG, MACSYMA, etc.) to of- signal processing is needed for progress in
For example, certain optimization-based de- fer a superior medium for design problem some veIy demanding areas, such as multi-
sign algorithms work only because of the definition, conceptualization, and imple- sensor feedback control of advanced robotic
human interaction via graphics (an imple- mentation. Serious research in new lan- manipulators. In addition. the limitations of
mentation of an algorithm question). guages and new architectures is needed here, VLSI chips must be properly understood in
A point that should be emphasized here is since these tools must be fairly flexible. de- relation to control systems design, and proper
that paying more attention to properly mod- scriptive. and fast. feedback should be provided to the manu-
eling the system and the controller often re- facturers.
Workstations
quires more sophisticated and new analytical
and computational tools. For example, it ap- In my opinion, what is needed is a system Finally, I would like to state that these
pears that although we are convinced that engineer’sworkstation, which is really a technological advances offer the possibility
non-Gaussian real-time estimation problems design superworkstation combining an AI of integrating design for many engineering
require massively parallel computations, the machine with a graphics engine and a multi- automation and signal processing systems. I
signal and observations models we use in processor “number cruncher.” This super- am confident that we will succeed in meeting
control systems have not been selected on workstation is part of a network of similar the challenges described here and that we
the basis of their ability to induce parallel and other workstations and computers, so will prosper from the opportunities they im-
computer implementations. New models are that the engineer can really have the capa- Ply.

8 lEEE Confro! Systems Magazine