Lapas attēli
PDF
ePub

Miscellaneous

A Look at Pulsed Power Research in Japan

M. Kristiansen

Observations from a recent visit to several Japanese industrial companies and
organizations involved in pulsed power (especially rail gun) research are
summarized.

Ocean Science

The First Asia-Pacific Conference on Algal Biotechnology
Aharon Gibor

This article presents recent research in three major areas of algal biotechnology:
large sea weeds, their products, and methods of cultivation; technologies for
cultivating microalgae and the products of these algae; and use of algal
technologies for solving ecological and environmental problems.

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
sea trials: a semi-autonomous vehicle and a new, low-light-level camera for a
remotely operated vehicle.

20th U.S.-Japan Joint Meeting: Sea Bottom Surveys Panel Pat Wilde

The reports and discussions from this meeting demonstrate the close cooperation
between the statutory seafloor mapping agencies of both countries and their
commitment to incorporate the latest advances in electronics and satellite
navigation to aid mariners and scientists alike.

Page

75

81

85

91

Page

Robotics

The Scoop on Ultrasonic Motors in Japan

Anita Flynn

This article describes the state of the art in ultrasonic motors in Japan.

Tele-Existence Work at the Research Center for Advanced Science and
Technology at the University of Tokyo

David K. Kahaner

Prof. S. Tachi's work on tele-existence is described.

Japanese Research in Intelligent Autonomous Robot Control

Yutaka Kanayama

This article reviews the Intelligent Robots and Systems Workshop and describes
related research presented at site visits.

[blocks in formation]

Cover: Yamabico-11, the robot designed by Prof. Yutaka Kanayama, Naval Postgraduate School (see his article on page 107).

UPDATE ON AN UNMANNED HELICOPTER WITH FUZZY CONTROLLER

Last year I wrote about helicopter flight control based on a fuzzy logic developed by Prof. M. Sugeno at the Tokyo Institute of Technology [D.K. Kahaner and D.G. Schwartz, "Fuzzy helicopter flight control," Scientific Information Bulletin 16(4), 13-15 (1991)]. Since then Prof. Sugeno has sent me some additional written material as well as several striking photos of the model helicopter in operation.

The project is to develop an unmanned helicopter for operation over water that would respond to simple voice controls such as "hover," "land," "fly straight," "turn left," etc.

The current system has 13 measured inputs, 3 angles of rotation--roll, pitch, and yaw--3 angular velocities, velocities and accelerations along three axes, and altitude. Two additional state variables (horizontal position) cannot be measured. Sensors are made by Tokimec; angles, angular velocities, and accelerations are measured by TMOS1000, the altitude by an ultrasonic wave sensor, and velocities by a microwave Doppler interference sensor. Global positioning information has not yet been incorporated, but plans are to include this when it becomes available here in Japan.

There are four outputs that are the four control inputs to the helicopter such as elevator adjustment for forwardbackward movement, aileron adjustment for left-right movement, throttle adjustment for up-down movement, and rudder adjustment for nose

direction.

SIBRIEFS

Scientific Information Briefs

The helicopter used in Sugeno's experiments is a Yamaha R-50, about 3.5 meters long with a 90-kg payload and a 98-cc engine.

The controller is based on fuzzy logic and is installed in a 16-bit microprocessor with a fuzzy inference engine built by Omron.

The most interesting aspect of the system is its hierarchical structure. At the top level is a navigator system that receives operator's instructions (hover, land, etc.) along with the present flight states of the helicopter from the sensors. The navigator provides as output both trim information (an equilibrium position of the helicopter's attitude) and the desired values of the control inputs. Both sets of information are input to the lower level, the stabilizer level, which is a servo system with the trim as its reference signal. The stabilizer consists of blocks corresponding to its flight modes, e.g., a sideways flight block, forward flight block, hover flight block, etc. Each block consists of four modules corresponding to the four control inputs (elevator, aileron, rudder, and throttle).

This kind of hierarchical and modular structure simplifies the acquisition of control rules as well as the controller design. It is also natural as a pilot recognizes that his own ability at control is also hierarchical. To improve stability and the control dynamics, there is also built-in feed-forward control action.

The navigator has two subsystems: a major one outputs standard trim and a minor one compensates it. The major subsystem has rules such as “If flight mode is HOVER and flight state is

[blocks in formation]

etc. Here EP=Pr-P (previous minus current pitch angle), dP is pitch angular velocity, and Eo is elevator output of the stabilizer.

Currently, the helicopter can support the following flight modes: hovering, hovering turn, forward/rearward flight, left/rightward flight, and stop. Other flight modes are in progress: takeoff, acceleration/deceleration, left/ right turn, climb/descent, and landing.-David K. Kahaner, ONRASIA

JANUARY 1992 WAVELET SEMINAR

The Audio Visual Information Research Group (AVIRG), or the Shichokaku Joho Kenkyu Kai, sponsors monthly seminars on a variety of scientific and engineering topics at the School of Engineering, Tokyo University, Hongo Branch. Attendance is open to the general public; there are no fees or no membership requirements. Twice a year one of the monthly seminars covers a subject in great detail in which a number of experts are invited to present a series of lectures. This particular seminar on wavelets was chosen to be one of the semi-annual big events for 1992. The topic was suggested by Hideki Kawahara of NTT Basic Research Laboratories.

The first lecture, "Wavelets (Theory)," was given by Masaaki Sato, from the ATR Auditory and Visual Perception Laboratories. Sato began his lecture with the astonishing statement that he does not understand the sudden and, perhaps, unjustifiably high interest in wavelets; none of the ideas associated with wavelets is entirely new, and none of the applications has produced results that are better than or cannot be produced by more conventional methods. He half-jokingly suggested that perhaps we had all gathered to check up on what is happening on

wavelets and are anxious lest we miss out on a potentially lucrative or imaginative application of wavelets. Perhaps so, as it appeared that the majority of the audience consisted of industrial scientists from large, corporate, basic research laboratories, e.g. NEC, NTT, Hitachi, IBM, HP.

The basic properties of wavelets such as their generation from a mother function using dilations and translations were introduced. Sato's presentation on the Fourier-wavelet analogy was clear and insightful. The timefrequency limitations of the Fourier methods were noted. In Fourier methods, the size of the sampling window is fixed, whereas in the wavelet method the window size can be set to short windows at high frequencies and long windows at low frequencies. The flexibility in choosing the sampling window size is a nice property; however, Sato favored the idea that much of the hoopla over wavelet expansions arises from the overcompleteness property of wavelet bases, a desirable but not particularly unique property. He discussed and later reiterated that Prof. Ogawa and his colleagues at To-ko-dai have been studying the role of overcompleteness properties and their applications to noise suppression in signal analysis. Discussion on the wavelet transform and its invertibility properties followed what has been published in the literature. Work on wavelets and Cantor sets by Arneodo, Grasseau, and Holschneider (1988) was briefly mentioned as time ran out. For further information, please contact:

Masaaki Sato

ATR Auditory and Visual

Perception Laboratories Sanpeidani, Inuidani, Seika-cho Soraku-gun, Kyoto 619-02, Japan

The second lecture, "Signal Processing of Human Voice Data," was given by Toshio Irino, NTT Basic

Research Laboratories. His talk was strikingly different in tone from that of Kawahara given at the Society for Industrial and Applied Mathematics (SIAM) wavelet seminars late last year. Irino's frankness was refreshing and should be commended. He discussed through demos how his initial optimism regarding possible applications of waveletbased techniques to voice compression and reconstruction was overly naive at best and entirely off-track at worst. It appears, at least for now, that results from his wavelet-based processing techniques cannot even equal those from more conventional methods. Perhaps I am wrong, but Irino's demo seemed to be the same as that given by Kawahara. It struck me as peculiar that two coauthors and joint researchers could set such different outlooks (optimistic versus pessimistic) on the future of their wavelet-based voice processing work. For further information, please

contact:

Toshio Irino

NTT Basic Research Laboratories 3-9-11 Midori-cho

Musashino-shi, Tokyo 180, Japan

Tel: +81-422-59-4201
Fax: +81-422-59-3393
E-mail: irino@nttlab.ntt.jp

Hideki Kawahara (same address as
Irino above)

Tel: +81-422-59-2276
E-mail: kawahara@nttlab.ntt.jp

Mutsumi Ohta (NEC C&C System Research Laboratories) gave the third lecture titled "Two-Dimensional Image Processing." He presented a brief sketch of the history of two-dimensional image processing to give the audience perspective on the evolution as well as current status of wavelets and their applications. He distributed a time chart comparing and pointing out the relationships between five processing methods: transform coding, subband

coding, pyramid coding, wavelet, and pattern recognition. Ohta believes that the conjugate quadrature filter (CQF) is the first method that can be called wavelet-like. He noted that Barnel's assumptions are similar to those of Mallat and Daubechies, but the latter have added conditions such as admissibility for mathematical rigor. Unfortunately, Ohta had incorrectly anticipated that Sato would outline the concept of multiresolution analysis, so an improvised summary was given.

Ohta noted that one of the more promising areas for applying wavelet techniques is in two-dimensional image processing, with time as a third dimension. Clean processing of rapidly moving objects is very difficult using discrete cosine transform (DCT) techniques. It is difficult to prevent severe blurring of the images; this blurring has been given many nicknames, such as the "mosquito effect" (for its similarity to the flapping of the wings) or the "corona effect." Ohta showed some results from his experiments in waveletbased image processing before concluding his talk. For further information, please contact:

Mutsumi Ohta

NEC C&C System Research

Laboratories

Terminal System Kenkyubu 1-1, Miyazaki 4-chome, Miyamae-ku

Kawasaki, Kanagawa 213, Japan

Shigeru Muraki's (Denshi Gijutsu Sogo Kenkyujo) talk on "Processing of Three-Dimensional Volumetric Data" consisted primarily of his own work. His first slide showed his work on resurrecting a three-dimensional "blobby" clay-like model from a two-dimensional slide. This type of computation-intensive, threedimensional imaging work led Muraki to be interested in wavelet techniques. He then gave a brief explanation of the

three-dimensional equations for wavelet expansions along with the data storage scheme to be used. In a second example, reconstructions of Dr. Spock's threedimensional facial features were shown for six different resolution levels where 512, 1,024, 2,048, 4,096, 8,192, and 16,384 functions were used. A final example showed how the technique could be applied to sets of two-dimensional magnetic resonance imaging (MRI) slices, with ray tracing to give a nearrealistic three-dimensional image. The presentation concluded with results from experiments where 2,048, 8,192, and experiments where 2,048, 8,192, and 32,768 functions were used and a vertical planar cut was made through the head to reveal the interior of the skull for each resolution level. For further information, please contact

Shigeru Muraki

Denshi Gijutsu Sogo Kenkyujo 1-1-4 Ume-sono Tsukuba-shi 305, Japan

--Mei Kobayashi, IBM Japan

INSTITUTE FOR NEW GENERATION COMPUTER TECHNOLOGY'S (ICOT) KL1 "ENGINE"

My most recent report on ICOT [“Japan's Fifth Generation Computer Project," Scientific Information Bulletin 16(3), 31-35 (1991)] described many of the project's goals and expectations. the project's goals and expectations. Please refer to that for more details.

ICOT is trying to develop hardware, called PIM (parallel inference machine), which will run programs written in their language, called KL1 (Knowledge Language). The targeted final system is called the Fifth Generation prototype system and is expected to be about 1,000 processors along with their associated system software. This system is intended to demonstrate the usefulness

of logic programming and its potential uses in artificial intelligence, databases, business tools, etc. The developers also specifically list numeric and scientific applications, although these seem far down among the areas that they are working on.

The KL1 language was developed as an extension of another logic programming language, GHC (Guarded Horn Clause). It is a kernel language, i.e., application programs as well as system software can be developed. Thus, it has goal reduction and goal scheduling, memory management, interrupt handling, input/output (I/O) processing, compiling, etc. Also, it has various parallel execution control functions such as pragma, distributed load control, priority control, and meta-control. In fact, the functions required are so complex that ICOT scientists have decided to use an intermediate language (KL1-B). Thus, users write in KL1, which is compiled into KL1-B for an abstract machine. A runtime system is designed for virtual hardware, which models a tag architecture and multiprocessor shared memory. This approach eases portability and program readability; only the low-level hardware dependent part needs to be altered for any real machine.

The fundamental driving features for the implementation of KL1 into a real parallel computer are that as many processors as possible should be kept busy, few processors should be wasted, and compatibility with software should be kept. ICOT's Keiji Hirata, in a recent paper ["Research and development of KL1 engine,” ICOT Journal 32, 2-13 (1991)], comments: "What kind of hardware architecture can support such an efficient parallel execution? We can find no proper answer to this question presently." Nevertheless, ICOT's researchers have generally agreed that the MIMD (multiple instruction, multiple data) model is most appropriate. Consequently, ICOT is

« iepriekšējāTurpināt »