Abstract
〈Vol.2 No.2(2009.3)〉
Titles
▲
■ Finite Time Control with Observer-Based Output Feedback \\ for Linear Discrete-Time Systems
KyuTech・Hiroyuki ICHIHARA and Shizuoka Univ.・Hitoshi KATAYAMA
In this paper we consider finite time stabilization and finite time boundedness
control problems for time-varying discrete-time systems. We give a set
of sufficient conditions, in terms of difference LMIs, for the existence
of observer-based output feedback controllers that make the system finite
time stable and finite time bounded. We then reduce the obtained results
to the ones for time-invariant discrete-time systems and derive numerically
tractable sufficient conditions given by LMIs. We also show numerical examples
to illustrate the design methods of observer-based output feedback controllers.
▲
■ Design Method of Fault Detector for Injection Unit
The Japan Steel Works, LTD.・Kiyoshi OCHI and Hiroshima University・Masami
SAEKI
An injection unit is considered as a speed control system utilizing a reaction-force
sensor.Our purpose is to design a fault detector that detects and isolates
actuator and sensor faults given that the system is disturbed by a reaction
force.First described is the fault detector's general structure.
In this system, a disturbance observer that estimates the reaction force
is designed for the speed control system in order to obtain the residual
signals, and then post-filters that separate the specific frequency elements
from the residual signals are applied in order to generate the decision
signals. Next, we describe a fault detector designed specifically for a
model of the injection unit.It is shown that the disturbance imposed on
the decision variablescan be made significantly small by appropriate adjustments
to the observer bandwidth,and that most of the sensor faults and actuator
faults can be detected and some of them can be isolated in the frequency
domain by setting the frequency characteristics of the post-filters appropriately.Our
result is verified by experiments for an actual injection unit.
▲
■ Grasp and Manipulation by Soft Finger with 3-Dimensional Deformation
Nagoya University・Akira NAKASHIMA,Takeshi SHIBATA and Yoshikazu HAYAKAWA
In this paper, we consider control of grasp and manipulation of an object
in 3-dimensional space by a 3-fingered hand robot with soft finger-tips.We
firstly propose a 3-dimensional deformation model of a semispherical soft
finger tip and verify its relevance by experimental data. Second, we derive
the contact kinematics and the dynamical equations of the fingers and the
object where the 3-dimensional deformation is considered. For the system,
we thirdly propose a method to regulate the object and the internal force
with the information of the hand, the object and the deformation. %We also
consider the case where the deformation and the object % information are
replaced by its estimated values from the hand information. A simulation
is shown to prove the effectiveness of the control method.
▲
■ Detection and Identification of an Inclined Crack in Concrete Structures
Using an Ultrasonic Sensor
Yamaguchi University・Shogo TANAKA and Muhammed MAZHARUL ISLAM
Conventional ultrasonic non-destructive inspection systems used for detecting cracks in concrete are based primarily on the visual inspection of the amplitude of the sensor output and therefore, are not highly reliable. Moreover, in an inclined crack, the point of reflection of the transmitted signal is not necessarily just below the sensor, because this point is determined by a trade-off between the directivity of the sensor and the tilt angle of the inclined crack. Therefore, the conventional methods are unreliable for identifying an inclined crack in concrete structures.
In this paper, we propose a simple and effective method to detect an inclined crack in concrete structures using an ultrasonic sensor with the signal propagation model and a method of least squares. First, the signal propagation model is used to obtain the propagation times of the reflected wave signals from the crack using a pattern matching technique and then a method of least squares is applied to identify the inclined crack accurately considering both the reflection efficiency and the signal strength at the point of reflection on the crack. Finally, experiments are performed to demonstrate the validity of the method.
▲
■ Development of SI Engine Control Education System
Waseda University・Dongmei WU,Masatoshi OGAWA,Harutoshi OGAI and Jin KUSAKA
TAn engine control education system is designed. This system can realize
the following functions: it serves to familiarize people with gasoline
engine properties, it can be applied to carry out engine control simulation,
to design engine control logic and to realize engine real-time simulation.
In the paper, the structure of this education system is explained. The
system is composed of a computer, a high-speed arithmetic processing board,
an ECU and an engine test bench. Engine control simulations are carried
out, and engine properties are obtained, therefore this system can assist
people in mastering gasoline engine properties. Besides, a real-time simulation
system is designed, and PID control real-time simulation is realized. In
the future, new control systems can be designed based on the current one.
When the engine simulator is connected with engine test bench and ECU,
engine real-time simulation can be realized.
▲
■ Guideline for Deciding Physical Parameters for Acrobot
California Institute of Technology・Atsushi BABA and Waseda University・Ryo
WATANABE
This paper presents a guideline for deciding physical parameters for the
Acrobot, a typical example of a two-link underactuated robot. A set of
physical parameters for the Acrobot are determined using an evaluation
function, which are derived by analyzing the robustness of an existing
swing-up controller for the Acrobot. A typical set of physical parameters
used in many papers are then evaluated using our evaluation function, and
the validity and performance of our proposed guideline are studied with
a series of simulations.
▲ ■ A Riemannian-Geometry Approach for Control of Robotic Systems under
Constraints
RIKEN-TRI Collaboration Center, Ritsumeikan University・Suguru ARIMOTO,RIKEN-TRI Collaboration Center・ Morio YOSHIDA,Ritsumeikan University・ Masahiro SEKIMOTO and Kyushu University・Kenji TAHARA
A Riemannian-geometry approach for control of two-dimensional object grasping and manipulation by using a pair of multi-joint planar robot fingers is presented, together with a basic discussion on stability of position and force hybrid control of redundant robotic systems under geometric constraints.
Even in the case that the shape of the object is arbitrary, it is possible to see that rolling contact constraints induce the Euler equation of motion in an implicit function form, in which constraint forces appear as wrench vectors affecting on the object.
The Riemannian metric can be introduced in a natural way on a constraint submanifold induced by rolling contacts.
A control signal called ``blind grasping" is defined and shown to be effective in stabilization of grasping without using the details of information of object shape and parameters or external sensing.
The concept of stability of the closed-loop system under constraints is renewed in order to overcome the degrees-of-freedom redundancy problem.
An extension of Dirichlet-Lagrange's stability theorem to a system of DOF-redundancy under constraints is presented by using a Morse-Lyapunov function.
▲
■ Combined Tactile Sensing with Optical Mouse Sensor
Jiangnan University,Saga University・Xin WANG,Saga University・Akira KIMOTO
and Katsunori SHIDA
Instead of functioning as part of a pointing device in computing, the optical
mouse sensor is employed for combined tactile sensing in this study. Based
on the experimental characteristics of the optical mouse sensor, with a
step-by-step table-look-up method, a combined tactile sensing system is
proposed. When an object surface is moved underneath the optical mouse
sensor, the translation direction, the category and the height of the object
surface can be obtained with the recorded output of the optical mouse sensor
based on a predefined data table.
▲ ■ Acquisition of Flexible Image Recognition\\ by Coupling of Reinforcement
Learning and a Neural Network
Oita University・Katsunari SHIBATA and Tomohiko KAWANO
The authors have proposed a very simple autonomous learning system consisting of one neural network (NN), whose inputs are raw sensor signals and whose outputs are directly passed to actuators as control signals, and which is trained by using reinforcement learning (RL).
However, the current opinion seems that such simple learning systems do
not actually work on complicated tasks in the real world. In this paper,
with a view to developing higher functions in robots,the authors bring
up the necessity to introduce autonomous learning of a massively parallel
and cohesively flexible system with massive inputs based on the consideration
of the difference in processing between humans and modern intelligent robots.
about the brain architecture and the sequential property of our consciousness.
The authors also bring up the necessity to place more importance on “optimization''
of the total system under a uniform criterion than “understandability''
for humans.
Thus, the authors attempt to stress the importance of their proposed system when considering the future research on robot intelligence. In this paper, by comparing with the processing in humans and robots, we introduce two viewpoints and %claims the usefullness of our proposed system again especially for developing higher functions hereafter. One viewpoint is the necessity of parallel architecture with large number of inputs, and the necessity of autonomous and cohesively flexible learning. The other is optimization of whole the process under an uniform criterion and emergence of purposive function based on the optimization. The information processing in the human brain is actually very sophisticated, but we are of the opinion that humans do not have enough ability to understand and express how the brain really works. The reason for this is that the human brain is a massively parallel and cohesively flexible system, while our consciousness seems sequential.
Nevertheless, many researchers seem to be trying to realize human-like intelligence as an extension of the present intelligent robots. Based on the discussion, this paper emphasizes the idea more than ever that in order to develop real intelligence in robots, the designer should leave everything to the autonomous learning ability in the couple of reinforcement learning (RL) and neural network (NN) rather than providing some process to robots. The couple of RL and NN has the potential to be a massively parallel and cohesively flexible system which we expect to develop higher functions.
The experimental result in a real-world-like environment shows that image
recognition from as many as 6240 visual signals can be acquired through
RL under various backgrounds and light conditions without providing any
knowledge about image processing or target object.
It works even for camera image inputs that were not experienced in learning.
In the hidden layer, template-like representation, division of roles between
hidden neurons, and representation to detect the target uninfluenced by
light condition or background were observed after learning.
The autonomous acquisition of such useful representations or functions
by a scalar reinforcement signal makes us feel the potential towards avoidance
of the frame problem and the development of higher functions. Interesting
representations were also observed in the hidden neurons, and that shows
the potential towards higher functions.
|
