« iepriekšējāTurpināt »
Technical Assessment of Two Unmanned Vehicles for Undersea Research
A.N. Kalvaitis and Gregory Stone
Two new Japanese developments in undersea technology were evaluated during
20th U.S. Japan Joint Meeting: Sea Bottom Surveys Panel
The reports and discussions from this meeting demonstrate the close cooperation
The Scoop on Ultrasonic Motors in Japan
This article describes the state of the art in ultrasonic motors in Japan.
Tele-Existence Work at the Research Center for Advanced Science and
David K. Kahaner
Prof. S. Tachi's work on tele-existence is described.
Japanese Research in Intelligent Autonomous Robot Control
This article reviews the Intelligent Robots and Systems Workshop and describes
Cover: Yamabico-11, the robot designed by Prof. Yutaka Kanayama, Naval Postgraduate School (see his article on page 107).
Scientific Information Briefs
UPDATE ON AN UNMANNED The helicopter used in Sugeno's FORWARD then standard trim is HELICOPTER WITH FUZZY experiments is a Yamaha R-50, about STRIM.” The latter is a four-tuple CONTROLLER
3.5 meters long with a 90-kg payload (P,E,R,A), which is the reference signals and a 98-cc engine.
of the pitch angle, roll angle, and the Last year I wrote about helicopter The controller is based on fuzzy offsets of the elevator and aileron. Since flight control based on a fuzzy logic logic and is installed in a 16-bit micro- true trim can deviate from the standard, developed by Prof. M. Sugeno at the processor with a fuzzy inference engine
processor with a fuzzy inference engine it must be compensated, given by Tokyo Institute of Technology built by Omron.
TRIM = STRIM + cTRIM. (D.K. Kahaner and D.G. Schwartz, The most interesting aspect of the CTRIM=PcTRIM+dcTRIM, where “Fuzzy helicopter flight control,” system is its hierarchical structure. At PCTRIM is the prior cTRIM and Scientific Information Bulletin 16(4), the top level is a navigator system that dcTRIM is output of the minor system. 13-15 (1991)]. Since then Prof. Sugeno receives operator's instructions (hover, The minor subsystem has rules as follows: has sent me some additional written land, etc.) along with the present flight material as well as several striking photos states of the helicopter from the sen- If X is + and dX is + then dcTRIM is v1 of the model helicopter in operation. sors. The navigator provides as output If X is + and dx is 0 then dcTRIM is v2 The project is to develop an
both trim information (an equilibrium unmanned helicopter for operation over position of the helicopter's attitude) water that would respond to simple and the desired values of the control If Y is - and dy is - then dcTRIM is v9 voice controls such as “hover,” “land,” inputs. Both sets of information are "fly straight," "turn left," etc.
input to the lower level, the stabilizer (Here X, Y are longitudinal and lateral The current system has 13 measured level, which is a servo system with the velocity and dX, DY are corresponding inputs, 3 angles of rotation--roll, pitch, trim as its reference signal. The stabi- accelerations.) and yaw--3 angular velocities, veloci- lizer consists of blocks corresponding The control rules in the stabilizer ties and accelerations along three axes, to its flight modes, e.g., a sideways were obtained firstly by interviewing and altitude. Two additional state flight block, forward flight block, hover pilots and then tested and modified variables (horizontal position) cannot flight block, etc. Each block consists of experimentally with a helicopter flight be measured. Sensors are made by four modules corresponding to the four simulator at Kawasaki Heavy IndusTokimec; angles, angular velocities, and control inputs (elevator, aileron, rud- tries. In total 54 rules were developed accelerations are measured by TMOS- der, and throttle).
based on pilot knowledge. Such knowl1000, the altitude by an ultrasonic This kind of hierarchical and mod- edge was formulated in statements like wave sensor, and velocities by a ular structure simplifies the acquisimicrowave Doppler interference sensor. tion of control rules as well as the "If the body pitches then control the Global positioning information has not controller design. It is also natural as a elevator in reverse" yet been incorporated, but plans are to pilot recognizes that his own ability at "If the body moves sideways, then control include this when it becomes available control is also hierarchical. To improve
the aileron in reverse" here in Japan.
stability and the control dynamics, there There are four outputs that are the is also built-in feed-forward control Each developed rule has two inputs four control inputs to the helicopter action.
and one output. Inputs are linguistic such as elevator adjustment for forward- The navigator has two subsystems: variables (positive, negative, zero). For backward movement, aileron adjust- a major one outputs standard trim and example, some elevator rules are ment for left-right movement, throttle a minor one compensates it. The major adjustment for up-down movement, subsystem has rules such as “If flight If EP is + and dP is + then Eoral and rudder adjustment for nose mode is HOVER and flight state is If EP is + and dP is ) then Eo=a2 direction.
etc. Here EP=Pr-P (previous minus wavelets and are anxious lest we miss Research Laboratories. His talk was current pitch angle), dP is pitch angular out on a potentially lucrative or imag
out on a potentially lucrative or imag- strikingly different in tone from that of velocity, and Eo is elevator output of inative application of wavelets. Per- Kawahara given at the Society for Industhe stabilizer.
haps so, as it appeared that the major- trial and Applied Mathematics (SIAM) Currently, the helicopter can sup- ity of the audience consisted of indus- wavelet seminars late last year. Irino's port the following flight modes: hover- trial scientists from large, corporate, frankness was refreshing and should ing, hovering turn, forward/rearward basic research laboratories, e.g. NEC,
basic research laboratories, e.g. NEC, be commended. He discussed through flight, left/rightward flight, and stop. NTT, Hitachi, IBM, HP.
demos how his initial optimism regardOther flight modes are in progress: The basic properties of wavelets such ing possible applications of wavelettakeoff, acceleration/deceleration, left/ as their generation from a mother based techniques to voice compression right turn, climb/descent, and landing.-- function using dilations and transla- and reconstruction was overly naive at David K. Kahaner, ONRASIA
tions were introduced. Sato's presen- best and entirely off-track at worst. It tation on the Fourier-wavelet analogy appears,
appears, at least for now, that results was clear and insightful. The time- from his wavelet-based processing tech
frequency limitations of the Fourier niques cannot even equal those from JANUARY 1992
methods were noted. In Fourier meth- more conventional methods. Perhaps I WAVELET SEMINAR
ods, the size of the sampling window is am wrong, but Irino's demo seemed to
fixed, whereas in the wavelet method be the same as that given by Kawahara. The Audio Visual Information the window size can be set to short It struck me as peculiar that two Research Group (AVIRG), or the windows at high frequencies and long coauthors and joint researchers could Shichokaku Joho Kenkyu Kai, spon- windows at low frequencies. The flexi- set such different outlooks (optimistic sors monthly seminars on a variety of bility in choosing the sampling window versus pessimistic) on the future of scientific and engineering topics at the size is a nice property; however, Sato their wavelet-based voice processing School of Engineering, Tokyo Univer- favored the idea that much of the hoopla work. For further information, please sity, Hongo Branch. Attendance is open over wavelet expansions arises from contact: to the general public; there are no fees the overcompleteness property of or no membership requirements. Twice wavelet bases, a desirable but not par- Toshio Irino a year one of the monthly seminars ticularly unique property. He discussed
NTT Basic Research Laboratories covers a subject in great detail in which and later reiterated that Prof. Ogawa 3-9-11 Midori-cho a number of experts are invited to pre- and his colleagues at To-ko-dai have Musashino-shi, Tokyo 180, Japan sent a series of lectures. This particular been studying the role of overcomplete- Tel: +81-422-59-4201 seminar on wavelets was chosen to be ness properties and their applications Fax: +81-422-59-3393 one of the semi-annual big events for to noise suppression in signal analysis. E-mail: email@example.com 1992. The topic was suggested by Hideki Discussion on the wavelet transform Kawahara of NTT Basic Research and its invertibility properties followed Hideki Kawahara (same address as Laboratories. what has been published in the litera
Irino above) The first lecture, “Wavelets ture. Work on wavelets and Cantor Tel: +81-422-59-2276 (Theory)," was given by Masaaki Sato, sets by Arneodo, Grasseau, and E-mail: firstname.lastname@example.org from the ATR Auditory and Visual Holschneider (1988) was briefly menPerception Laboratories. Sato began tioned as time ran out. For further Mutsumi Ohta (NEC C&C System his lecture with the astonishing state- information, please contact:
Research Laboratories) gave the third ment that he does not understand the
lecture titled "Two-Dimensional Image sudden and, perhaps, unjustifiably high Masaaki Sato
Processing." He presented a brief sketch interest in wavelets; none of the ideas ATR Auditory and Visual
of the history of two-dimensional image associated with wavelets is entirely new, Perception Laboratories processing to give the audience perand none of the applications has pro- Sanpeidani, Inuidani, Seika-cho spective on the evolution as well as duced results that are better than or Soraku-gun, Kyoto 619-02, Japan current status of wavelets and their cannot be produced by more conven
applications. He distributed a time chart tional methods. He half-jokingly sug- The second lecture, “Signal Pro- comparing and pointing out the relagested that perhaps we had all gathered cessing of Human Voice Data,” was tionships between five processing to check up on what is happening on given by Toshio Irino, NTT Basic methods: transform coding, subband coding, pyramid coding, wavelet, and three-dimensional equations for wavelet of logic programming and its potential pattern recognition. Ohta believes that expansions along with the data storage uses in artificial intelligence, databases, the conjugate quadrature filter (CQF) scheme to be used. In a second example, business tools, etc. The developers also is the first method that can be called reconstructions of Dr. Spock's three- specifically list numeric and scientific wavelet-like. He noted that Barnel's dimensional facial features were shown applications, although these seem far assumptions are similar to those of for six different resolution levels where down
among the areas that they are Mallat and Daubechies, but the latter 512, 1,024, 2,048, 4,096, 8,192, and 16,384 working on. have added conditions such as admissi- functions were used. A final example The KL1 language was developed bility for mathematical rigor. Unfortu- showed how the technique could be as an extension of another logic pronately, Ohta had incorrectly anticipated applied to sets of two-dimensional gramming language, GHC (Guarded that Sato would outline the concept of magnetic resonance imaging (MRI)
magnetic resonance imaging (MRI) Horn Clause). It is a kernel language, multiresolution analysis, so an impro- slices, with ray tracing to give a near- i.e., application programs as well as vised summary was given.
realistic three-dimensionalimage. The system software can be developed. Thus, Ohta noted that one of the more presentation concluded with results from it has goal reduction and goal schedulpromising areas for applying wavelet experiments where 2,048, 8,192, and ing, memory management, interrupt techniques is in two-dimensional image 32,768 functions were used and a verti- handling, input/output (I/O) processprocessing, with time as a third dimen- cal planar cut was made through the ing, compiling, etc. Also, it has various sion. Clean processing of rapidly mov- head to reveal the interior of the skull parallel execution control functions such ing objects is very difficult using dis- for each resolution level. For further as pragma, distributed load control, crete cosine transform (DCT) tech- information, please contact
priority control, and meta-control. In niques. It is difficult to prevent severe
fact, the functions required are so blurring of the images, this blurring Shigeru Muraki
complex that ICOT scientists have has been given many nicknames, such Denshi Gijutsu Sogo Kenkyujo decided to use an intermediate lanas the “mosquito effect" (for its simi- 1-1-4 Ume-sono
guage (KL1-B). Thus, users write in larity to the flapping of the wings) or Tsukuba-shi 305, Japan
KL1, which is compiled into KL1-B for the “corona effect.” Ohta showed some
an abstract machine. A runtime system results from his experiments in wavelet- --Mei Kobayashi, IBM Japan
is designed for virtual hardware, which based image processing before conclud
models a tag architecture and multiing his talk. For further information,
processor shared memory. This please contact:
approach eases portability and program INSTITUTE FOR NEW
readability; only the low-level hardMutsumi Ohta
GENERATION COMPUTER ware dependent part needs to be altered NEC C&C System Research TECHNOLOGY'S (ICOT) for any real machine. Laboratories KL1 "ENGINE"
The fundamental driving features Terminal System Kenkyubu
for the implementation of KL1 into a 1-1, Miyazaki 4-chome,
My most recent report on ICOT real parallel computer are that as many Miyamae-ku
(“Japan's Fifth Generation Computer processors as possible should be kept Kawasaki, Kanagawa 213, Japan Project,” Scientific Information Bulletin busy, few processors should be wasted,
16(3), 31-35 (1991)] described many of and compatibility with software should Shigeru Muraki's (Denshi Gijutsu the project's goals and expectations. be kept. ICOT's Keiji Hirata, in a recent Sogo Kenkyujo) talk on “Processing of Please refer to that for more details.
. paper (“Research and development of Three-Dimensional Volumetric Data" ICOT is trying to develop hardware, KL1 engine,” ICOT Journal 32, 2-13 consisted primarily of his own work. called PIM (parallel inference machine), (1991)), comments: “What kind of His first slide showed his work on which will run programs written in their hardware architecture can support resurrecting a three-dimensional
three-dimensional language, called KL1 (Knowledge such an efficient parallel execution? “blobby” clay-like model from a Language). The targeted final system We can find no proper answer to this two-dimensional slide. This type is called the Fifth Generation prototype question presently.” Nevertheless, of computation-intensive, three- system and is expected to be about ICOT's researchers have generally dimensional imaging work led Muraki 1,000 processors along with their asso- agreed that the MIMD (multiple instructo be interested in wavelet techniques. ciated system software. This system is tion, multiple data) model is most He then gave a brief explanation of the intended to demonstrate the usefulness appropriate. Consequently, ICOT is