HISTORICAL PERSPECTIVES

My Experiences in the Control Field

JUERGEN E. ACKERMANN

n 1956, I had to make an important decision: What shall I study? My favorite subjects were architecture and electrical engineering. In either field I hoped to design something newnot only analyze existing phenomena. I still admire interesting architecture like the Sydney Opera House, as well as the Olympic Park and the new BMW World in Munich. On the other hand, I was a radio amateur, and that was the deciding factor in favor of communication engineering. I chose Darmstadt, Germany, because Karl Kuepfmueller was a professor there.

I looked for a job that would give me the opportunity to study in the United States and offer challenging applications. That was aerospace.

BECOMING A CONTROL SCIENTIST UNDER THE GUIDANCE OF ELI JURY

After completing my studies, I looked for a job that would give me the opportunity to study in the United States and offer challenging applications. That was aerospace. George Knausenberger at DLR (the German aerospace research organization) in Oberpfaffenhofen had the necessary connections to provide me with a NASA University Fellowship and with some German government support. I chose the University of California at

BECOMING A CONTROL ENGINEER UNDER THE GUIDANCE OF WINFRIED OPPELT

Then Winfried Oppelt came as a new professor with a new subject: control engineering. In his first lecture he talked about feedback and stability. The examples came from different engineering fields, from economy, and from biology. I found it fascinating that common principles applied throughout. I chose this subject because it did not force me into an early specialization. Oppelts research focus was on describing functions, and for my first thesis he gave me the topic of a describing function for the sampler and hold. In my doctoral thesis I later generalized this approach to nonlinear sampled-data systems. Oppelt gave me both bad and good advice. The bad one was: publish in the German control journal Regelungstechnik. The good one was: learn more mathematics. Go to a U.S. university for further studies.
Digital Object Identifier 10.1109/MCS.2008.929900

Berkeley because I knew a book by Eli Jury, then at Berkeley, from my sampled-data studies. At that time (1963) Charles Desoer and Lotfi Zadeh were writing their book Linear System Theory, and we were learning from the draft chapters. The state-space theory was completely new to me, although I had a German diploma in control engineering. For my M.S. thesis I worked with Eli Jury on stability criteria for discrete-time systems. I was interested in the redundancy in algebraic criteria and learned about a 1929 paper

Illinois, Toronto, and Oberpfaffenhofen friends: Bill Perkins, Murray Wonham, Joe Cruz, Juergen Ackermann, and Petar Kokotovic, Boston 1975.

DECEMBER 2008

IEEE CONTROL SYSTEMS MAGAZINE 115

Authorized licensed use limited to: University of Michigan Library. Downloaded on November 25, 2008 at 19:09 from IEEE Xplore. Restrictions apply.

by Frazer and Duncan, where they develop the boundary crossing idea: the stable neighborhood of a stable point in the coefficient space of a fixed degree polynomial is bounded by the condition that the last Hurwitz determinant vanishes. I exploited this concept in my later work on robust control.

RESEARCH AT DLR
Initially we had a lot of freedom in our research at DLR, and we were generously funded by the German Federal Department of Research and Technology. Under the leadership of Gerd Schneider, a team of colleagues, who had also spent a year at U.S. universities, built up a pool of knowledge in modern control. Also, the growing aerospace industry in Germany noted that the traditional training of control engineers in Nyquist plots and root loci was not sufficient to tackle challenging aerospace problems. So they sent many practicing engineers to our continuing education courses organized by the Carl-Cranz-Gesellschaft (CCG), cofounded by George Knausenberger. My part was Sampled-Data Control Systems. The lecture notes were the starting point for my book on that topic, published in 1972. It also had a section Pole Placement Without Canonical Form, which later became known as Ackermanns formula. This formula gives an understanding of what happens in terms of feedback gains if you shake a real closed-loop eigenvalue or a complex pair of them. In a later 1985 English edition of the book, the multivariable generalization and numerical stability by Hessenberg transformation was also given. Yakov Tsypkin proposed that the first book be translated to Russian, but he was asked by the officials: Is there no scientist in the Soviet Union who can write such book? That killed the project, but Yakov arranged for a Polish translation. Further translations to Chinese and English followed. My first practical project was a low-cost, star-oriented spinning rock-

et tip to be controlled by a single gas jet. I formulated the theory of this controlled gyro and developed a graphical solution approach for the on-off controller design problem, such that both precession and nutation were damped simultaneously and fast during reorientation. Our DLR institute at Oberpfaffenhofen became an attraction point for leading control scientists from all over the world. For example, Richard Bucy chose our institute for his sabbatical. Over a beer or three in Andechs, we discussed our topics on paper napkins: my multivariable deadbeat solution by assigning a minimal polynomial with all roots at zero, and his ideas about canonical forms with the minimum number of parameters and minimal realization in such form. I later continued this line of research with finite effect sequences for model verification. In the early 1970s our budgets became tight. Most members of our team accepted professorships at German universities or moved into industry positions. I declined such offers and was quickly promoted to management positions at DLR. The first one was head of the control department, and later I moved on to

the position of director of the institute. I attracted and supported the department heads Georg Gruebel, Gerd Hirzinger, Willi Kortuem, Richard Schwertassek, and Klaus Well. Together we initiated and guided many applied research projects, which included robust autopilots and jet engine control, trajectory optimization of aircraft, missiles and spacecraft, control of large flexible structures, robotics for space and manufacturing application, modeling and control of trains, maglev vehicles, and automobiles. We overcame our budget issues, but now our work had entirely different priorities ($$$). I was concerned that I did not have enough time for my personal research interests. I discussed this problem with Petar Kokotovic on a hike in the Rocky Mountains. He suggested that I come for a sabbatical to Urbana-Champaign. Initially, I thought that it would be completely impossible. However, the discussion with Petar helped me get my priorities in order. I had to solve some personnel questions first. In particular I had to choose my substitute at DLR during that year. Gerd Hirzinger did an excellent job, and I was free for the sabbatical.

Dick Bucy and Juergen Ackermann at the Automatic Control in Space Symposium in Toulouse 1970.

Petar Kokotovic contemplating Juergen Ackermanns decision between science and management, Banff Park 1977.

116 IEEE CONTROL SYSTEMS MAGAZINE

DECEMBER 2008

Authorized licensed use limited to: University of Michigan Library. Downloaded on November 25, 2008 at 19:09 from IEEE Xplore. Restrictions apply.

Vadim Utkin, Karl and Bia Astrom, Petar Kokotovic, and Eli Jury at Juergen Ackermanns 1987 IFAC Party in Herrsching.

Two discrete-time giants: Yakov Tsypkin and Eli Jury on their way to Juergen Ackermanns IFAC Party 1987. (Photo taken by Katsuhisa Furuta).

A NEW START ON ROBUST CONTROL IN ILLINOIS

Friends at Urbana-Champaign, in particular Joe Cruz, Petar Kokotovic, Bill Perkins, and Juraj Medanic, helped me and my wife Christa get started. It was a new experience to have our four kids in American schools. Another lesson: Get your drivers license first, and do not tell your car insurance agent that you will leave the United States after a year. I was lucky to find a masters student, Norm S. Franklin, who had worked in a control design team at McDonnell-Douglas after his bachelors degree, and now came back for his masters degree. He had worked on a gain-scheduled control system for an F4-E fighter aircraft, which was unstable due to the added canards. He knew the model data and specifications for normal operation and for sensor faults. When I told him that we would redesign the control system for fixed gains rather than gain scheduling, he probably thought I was crazy. That must be impossible, otherwise his Mc Donnell-Douglas team would have chosen this simpler solution. However, he accepted and mastered the challenge. I developed a parameter

space theory of robust control based on the above-mentioned boundary crossing concept, but keeping the frequency as a real parameter rather than eliminating it. Norm Franklin implemented the parameter space

and when was it published? I: In the German journal Regelungstechnik, six years ago. Student: Oh, if it is in German, then it is worthless. That was crude, but a good lesson for me. From then on I ignored Oppelts

Initially we had a lot of freedom in our research at DLR, and we were generously funded by the German Federal Department of Research and Technology.

approach for the military specifications of an admissible region for the eigenvalues of the controlled aircraft and for two identical sensor faults. We also demonstrated the general superiority of gyros over accelerometers in robust vehicle control. At the Allerton conference in 1978 a Chinese student of Charles Desoer presented my pole-placement formula without mentioning my name. After the session I went up to him and asked: Do you know my publication on this subject? Student: No, where

advice and published in English. Then I sent my publication of 1972 to Charles Desoer, and he answered that it was just a little student exercise and would not be published. A year later the book Linear Systems by Tom Kailath appeared. He knew my result from a visit to Oberpfaffenhofen and coined the name Ackermanns formula. So, finally, I received recognition in the English literature. One of the recognitions was that a student at an American Control Conference asked me:

DECEMBER 2008

IEEE CONTROL SYSTEMS MAGAZINE 117

Authorized licensed use limited to: University of Michigan Library. Downloaded on November 25, 2008 at 19:09 from IEEE Xplore. Restrictions apply.

Are all small tomato plants stable? Vladimir Kharitonov testing the lower and upper bound. Answer: some nonlinear ones need a stick. St. Petersburg, 1991.

Brian Anderson: Juergen, in which of your many capacities are you at this reception? Juergen Ackermann: In my drinking capacity, and there is more wine.

Are you the Ackermann of Ackermanns formula? When I said yes, he answered: I thought he must be long dead. Conversations like that keep you alive.

BACK IN OBERPFAFFENHOFEN
From now on robust control and applications were my topics in research and courses at the Technical University Munich (TUM) and CCG. One of the CCG participants was Wolfgang Darenberg of Daimler-Benz. He was working on automatic path tracking of a city bus. He noted that our robust flight control methods might help him find a solution. So, I had several TUM students work on parameter space controller designs for the bus and experimental verification on the Daimler-Benz test track in Rastatt. The robustness analysis was tough, because the velocity and the ratio mass/(friction coefficient between road and tire) enter polynomially into the coefficients of the closed-loop characteristic polynomial, but the mathematical background of Dieter Kaesbauer helped overcome this difficulty.

In 1981, I attended a meeting on mathematical theory of systems in Bielefeld. There I met Andrej Olbrot from Warsaw. During lunch he introduced me to Kharitonovs theorem on the back of an envelope. I tried to invite Vladimir Kharitonov to a robustness workshop, but he never received my letters. It was not until IFAC 1990 in Tallin that we met. I asked Andrej to present the result at a robust control workshop that I organized in Interlaken in 1981. Bob Barmish was the participant who immediately recognized the importance of Kharitonovs theorem. He made the concept popular in the West. Soon simpler proofs based on zero exclusion from the value set appeared. At a later robustness workshop, Brian Anderson told me during a break that Kris Hollot and his student Andy Bartlett had a further testing set result. It suffices that the edges of a parameter box are stable, if the parameters enter affinely in the polynomial coefficients. I invited Andy to Oberpfaffenhofen to be coauthor of my book on robust control published in 1993.

It was small talk like that in a network of peers that made the attendance at conferences and workshops so valuable. In this context I remember Neil Munroe telling me, during a break at the 1999 European Control Conference, about a result of Shankar Bhattacharyya and coworkers. In PID controller design with fixed Kp, the stable region in the (Kd-Ki)-plane consists of convex polygons. This observation motivated a new topic in our parameter space design of robust controllers, as the first step toward automated PID controller design. After the testing-set results for interval and affine polynomials, there was hope in the community that general nonlinear parameter dependence could be handled. Hanzhong Hu and I destroyed this hope by a frequently quoted benchmark example with an isolated unstable point. Nonlinear parameter dependence can be treated, however, if it appears in a tree-structured form such that some of the uncertain parameters appear only once in additive or multiplicative subpolynomials. Bob Barmish, myself,

118 IEEE CONTROL SYSTEMS MAGAZINE

DECEMBER 2008

Authorized licensed use limited to: University of Michigan Library. Downloaded on November 25, 2008 at 19:09 from IEEE Xplore. Restrictions apply.

In my retirement I particularly enjoy that I do not have to fight for budgets any more. I only do things that I am interested in.

and my student Wolfgang Sienel developed a method to discover tree structures and to exploit them in constructing value sets even for large numbers of parameters.

MAN-MACHINE INTERACTION
As an intermediate step toward fully automatic driving, I was interested in man-machine systems, where the driver or pilot takes care of the slow dynamics of path following, while the automatic control takes care of the fast

ROBUST CONTROL OF CARS AND AIRCRAFT

During my sabbatical at the University of California Irvine in 19891990, I taught a new course on robust control, and I used car steering as the example for a series of exercises. Gyros reduced the tracking errors for the track-following bus by at least 50%, but gyros were too expensive at that time for use in automobiles; nowadays yaw-rate sensors mass produced for ESP systems are cheap, and our control concepts from 1990 become practically feasible. My student Juergen Guldner demonstrated this in the PATH project at Berkeley in 1999. I developed the concept of structural unilateral decoupling of the lateral motion from the influence of the yaw motion. This concept works for a decoupling point near the front axle of a car and near the vertical fin of an aircraft. It is far superior to any numerical optimization of quantities at the center of gravity because it is structurally robust. In the IFAC Council I initiated the foundation of a Technical Committee on Automotive Control. I thought this would be a self-running proposal, but Lennart Ljung called me to let me know that unless I fought for it at the council meeting in 1991, my proposal would be dead. Obviously I had stepped into somebodys claim. Finally, my proposal was accepted, and today the committee is very successful.

dynamics like skidding or rollover of a car or an engine out during the start or landing of an aircraft. In cooperation with the Airbus structural dynamics division, Michael Kordt and I were able to reduce the loads of an aircraft structure and thereby its

Advice to Newcomers to the Field

ach career is unique, so you cannot follow general guidelines. Nevertheless, I provide some recommendations from my personal experience. Maybe one or the other advice will help young scientists or engineers. a) Publish your new results in English journals, like Automatica, IEEE Transactions on Automatic Control, IEEE Transactions on Control Systems Technology, or this magazine (IEEE Control Systems Magazine). You may get it published faster if you submit it to a conference first. b) Try to establish a worldwide network of researchers in your area of interest. Also talk to experts in other disciplines. c) Get your hands on application problems, and make use of the entire toolbox of the control engineer. If the only tool that you have is a hammer, then every problem looks like a nail. d) Come back later to your practical experience and analyze the problem, even if the project is finished. For example do not ask: Can we afford a gyro in this project, but ask: What would be the advantage of using a yaw-rate sensor? Typically you need an academic environment to follow this advice. Industry would hardly give you the time and permission to publish. e) Study control principles of problem classes like vehicles, robots, the human skeleton, and chemical processes rather than optimize something with the data you get. A symbolic solution for a class of plants is worth much more than a numerical one. f) I had good experience with putting up a conjecture and offering a prize for a proof or counterexample. g) Check carefully whether or not you really want to have the job of your boss. For a manager it is frequently difficult to go back to a lower paid scientist job with more academic freedom. h) Filing a patent is tedious and time consuming. You get inane objections to your idea from nonprofessionals and exaggerated expectations in terms of license income from your employer. You may be better off to sign a consultancy contract, but see that you keep publication rights, otherwise you are not recognized in the academic world. i) Do not believe in simulation results based on unverified model assumptions. j) Do not hesitate to step into somebodys claim if you are convinced that you have the better argument.

DECEMBER 2008

IEEE CONTROL SYSTEMS MAGAZINE 119

Authorized licensed use limited to: University of Michigan Library. Downloaded on November 25, 2008 at 19:09 from IEEE Xplore. Restrictions apply.

weight by the same control concept that worked so well for the car. I filed several patents on the car-steering results. But obviously it is difficult for an automotive company to admit that something crucial for the vehicle qualities was invented elsewhere. Also the high license fee expectations of my employer DLR were an obstacle.

In my retirement I particularly enjoy that I do not have to fight for budgets any more. I only do things that I am interested in.

AUTHOR BIOGRAPHY
Juergen E. Ackermann has been active as a control engineer, researcher, research manager, and educator, mostly in his native Germany, but including three sabbaticals at U.S. universities. From 1974 to 2001 he was director of the German Aerospace Center (DLR) Institute of Robotics and Mechatronics in Oberpfaffenhofen, Germany. He is currently an adjunct professor at the Technical University in Munich. A selection of his patents, books, plenary talks, and papers can be found at www.robotic.de/ Juergen.Ackermann. He is a Fellow of IEEE and IFAC. In 1996 he received the IFAC Nichols Medal and present-

AFTER RETIREMENT
In 2001 I retired from DLR because I had reached the age limit of 65 years. Gerd Hirzinger is my successor as director of the Institute of Robotics and Mechatronics. I continue to teach a course on robust control at the TUM and advise a few students with their thesis. I organized a DLR alumni group, where former directors of DLR meet socially and listen to talks on current DLR projects. In addition, I developed a series of talks on democracy for students.

ed the IEEE Bode Prize Lecture. He served the IEEE Control System Society as associate editor-at-large of IEEE Transactions on Automatic Control, as a member of the Board of Governors, and as chair and member of various committees. For IFAC he served on the council, the theory committee, and several awards committees. He was a member of the senate of DLR, the board of trustees of Deutsches Museum and of CCG, elected chief reviewer of Deutsche Forschungsgemeinschaft, and member of the editorial board of the journal Automatisierungstechnik. He was on the advisory board of the VDI/VDE Gesellschaft fuer Automatisierungstechnik and received a best publication award from this society. He is a recipient of the Johann-MariaBoykow award and received an OttoLilienthal Fellowship.

Hunting in the Deadzone

oulomb friction causes mechanisms to be resistant to moving from rest. A well-known phenomenon concerned with Coulomb friction is that the table will not start to move until the driving torque is large enough to break the static friction torque. Such characteristics of Coulomb friction form a deadzone nonlinearity in the system [12,13]. This deadzone results in the inactivity of position control systems to small errors if the integrator is not built in the controller. Integral action can eliminate the steady state error caused by the deadzone but may also lead the hunting problem. Hunting is an integral-induced stick-slip oscillation about the goal position when Coulomb friction is conspicuous in the mechanism [8,1416]. The cause of hunting can be stated as follows. From the Stribeck effect [2,8] or the study of Tustin [17], friction becomes larger at very low velocities. This feature of friction makes the table liable to stop before reaching the setting point. Once the table gets stuck, it will not move until breakaway occurs. The breakaway then reduces friction from its maximum static value down to its sliding level. Resultantly the over driven table slides across the goal position and has to reverse its velocity for going back to its target. The table thus comes to a standstill until the reversed error accumulates enough control effort to pull the table back. Then the scenario repeats. This stop-and-slide cycle causes the oscillations of the table around the desired position. R.-H. Wu and P.-C. Tung, Studies of Stick-Slip Friction, Presliding Displacement, and Hunting, ASME J. Dyn. Sys. Meas. Control, vol. 124, pp. 111117, 2002.

120 IEEE CONTROL SYSTEMS MAGAZINE

DECEMBER 2008

Authorized licensed use limited to: University of Michigan Library. Downloaded on November 25, 2008 at 19:09 from IEEE Xplore. Restrictions apply.