Wy 40 120 160 200 240 280 320 360 Feed along X-axis mm Figure 3: Straightness error motion of the probe along X axis COMPOSITION OF PLANELIKE ERROR FORM The error forms in Figure 3 are c
Trang 1Ch75-I044963.fm Page 374 Tuesday, August 1, 2006 5:58 PM Ch75-I044963.fm Page 374 Tuesday, August 1, 2006 5:58 PM
374
environment In case of the STPM the operation to evaluate the bundle of the straightness error forms
is easily realized by the CNC command of the machine
0.2
| 0 - 1
i °
LJ -0.1 -o.;
-o.:
/ ^ L X _ T T\ Max.
Wy
40 120 160 200 240 280 320 360
Feed along X-axis mm
Figure 3: Straightness error motion of the probe along X axis
COMPOSITION OF PLANELIKE ERROR FORM
The error forms in Figure 3 are composed of a bundle of the straightness error forms which aremeasured parallel to the X axis and are repeated stepwise to the direction of the Y axis
In coordinate system (X, Y, Z) among the work space of the CMM, a plane which is prescribed by (X,,
Yj, Zk) is considered, where i = 0, 1, ••• 1, j = 0, 1, ••• m, k = 0, 1, ••• n A bundle of the straightnesserror forms parallel to the X axis are measured on a plane of Zo, that is, (X;, Yj, Zo) In practicalapplication of the STPM, slight discrepancy A is inevitably left between the tip of two sensors(Tanaka & Sato (1986)) It has been demonstrated that iA is linearly accumulated at i-th position ofthe sensors on the measured straightness error form Then corrected error form can be obtained bysubtracting the following accumulation from the raw measured results,
Figure 4 illustrates the flow of composing a planelike error form mentioned above Zx and Zy indicatethe straightness errors of the probe for Z axis that are measured by applying the STPM along the Xand Y axes respectively
In Figure 5(a) a planelike error form for the probe movement is constructed from the straightness errorforms in Figure 4 The planelike error form for the objective Al plate can be simultaneously evaluated
in the procedure, if necessary, as shown in Figure 5(b)
ERROR SPACE AND ITS COMPENSATION
By the method proposed above, a planelike error form for arbitrary Zk can be constructed, and, it ispossible to estimate the error space by constructing enough number of planelike error forms Figure 6
Trang 20 50 100 150 200 250 300 350 400
0 50 100 150 200 250 300 350
axi
mm)
0 100 200 300 400
0 100 200 300 -15 -10 -5 0 5 10 15
illustrates an experimental example of detecting an error space for the Z axis The same procedure can
be applied to the directions X and Y which are vertical to Z, and planelike error forms for the Xi and
Yj are constructed Consequently, error space for 3 axes (X, Y, Z) is constructed Then it is possible toestimate error components for (X, Y, Z) at arbitrary points among the measurement range of a CMM.The points on which errors are directly measured are discrete The error at an arbitrary point can beobtained by applying interpolation or least squares method (LSM) When the error ez(X, Y, Z) isobtained by interpolation or LSM, the compensation of the error is easily performed by subtracting the
ez from Z By applying the method, the measurement accuracy is improved without any change ofhardware configuration
Measurement along the direction o f X axis for j=0 ••- m
Z x (Xj,Y|.Z 0 ) (i=0.-,];j=0,-,m)
Calculation of decrepancy A from Z X ( X | , Y 0 , Z Q )
A=Z X (X I ,Y,,,Z O )/I
Compensation for Z ^ X ^ Y J Z Q ) (i=0.-.l:j=l. .m)
SO that Z x (X ( ) Yj.Z 0 )=Z y (X 0 Yj,Z 0 ) ( j = l - m )
Verification of measurement
by comparing Z X ( X | Y J ;Z Q ) and Z (X h YyZQ) (j=0.-,m)
ZxpCl.Yo.Zo)
Figure 4: Flow of composing planelike error form for Z=Zo
Figure 5 (a) (left): Error curved surface of the straightness motion of the probe
(b) (right): Surface profile of the objective plane
Trang 30 50 100 150 250 350
0 50 100 200 300 350 -0.6 0 0.2 0.8
0 50 100 150 250 350 400
0 50 100 200 300 350 -1 -0.5 0 0.5
Z=40mm
0 50 100 200 300 400
0 50 100 200 300 -2 0 2 4
Fe X- mm) Feeda longY
measurement operation can be easily extended to planelike error form
An algorithm which constructs planelike error form from a bundle of the straightness error formswas demonstrated, in which a method that gives conditions at the both ends of the error forms isproposed
A procedure which derives error space for measurable range of the machine on the basis ofplanelike error forms is proposed This makes it possible to compensate error among the totalmeasurement range, which certainly improves the accuracy performance of the machine
ACKNOWLEDGEMENTS
The authors express their sincere thanks for the following support by Project "Advancement ofMeasurement of Macro Error Form (97S21-005)" from 1997 to 1999 by New Energy DevelopmentOrganization (NEDO), and Project "Advancement of Straightness Error Measurement by SequentialTwo Points Method" by Foundation, Promotion of Machine Tool Engineering in 1996 and 1997
REFERENCES
Balsamo A., Franke M., Trapet E., Waldele F., Jonge L.D and Vanherck, P (1997) Results of theCIRP-Euromet Intercomparison of Ball Plate-Based Techniques for Determining CMM Parametric
Errors, Annals CIRP 46:1, 463-466.
Evans C , Hocken R and Estler W.T (1996) Self-Calibration: Reversal, Redundancy, Error
Separation, and Absolute Testing, Annals CIRP 45:2, 617-634.
Kunzmann H and Waldele F (1988) Performance of CMMs, Annals CIRP 37:2, 633-640.
Kunzmann H., Trapet E and Waldele F (1995) Results of the International Comparison of Ball Plate
Measurements in CIRP and WECC, Annals CIRP 44:1, 479-482.
Tanaka H and Sato H (1986) Extensive Analysis and Development of Straightness Measurement by
Sequential Two-Points Method, Trans ASME, J Eng'glnd 108:3, 176-182.
Tozawa K., Sato H and O-hori M (1982) A New Method for the Measurement of the Straightness of
Machine Tools and Machined Work, Trans ASME, J Mack Des 104:3, 587-592.
Zhang G., Veale R., Charlton T., Borchardt B and Hocken R (1985) Error Compensation of
Coordinate Measuring Machines, Annals CIRP 34:1, 445-448.
Trang 4Terunao HIROTA1, Ryouji OKUTSU2, Kouji IZUMI2
V Inst of Environmental Studies, Graduate School of Frontier Science, The University of
Tokyo,7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan,
:/Yamatake Corporation, 4-1-1 Samukawa-Machi, Kohza-Gun, Kanagawa, 253-0113, Japan
ABSTRACT
In control valves, the sealing characteristics of gland packing has a close relation to the pressure of theradial direction of the packing, and tightening the packing in the direction of the axis controls themagnitude of the pressure in radial direction So far, measurement of this radius direction pressure hasbeen difficult, involving problems regarding the changes of the radial pressure over time or thedistribution of the pressure
Thus, in this paper, a method of measuring radial pressure through grease was proposed using a sensorfor measuring the pressure of a fluid, and measurement equipment based on this idea was produced.First, the validity of this measuring method was evaluated Next, the action of the piled-up packings in
a bonnet was clarified by applying this measuring method to the system
In order to prevent failure, it is necessary to clarify the behavior of the gland packing The sealingcharacteristics of gland packing are strongly affected by the magnitude of the contact pressure betweenpacking and the stem or the bonnet (=Radial Pressure), and this magnitude is controlled by tighteningthe pressure in the direction of the axis (=Axial Pressure) (Fig 1) Therefore, in order to prevent failure,
it is necessary to clarify the relation between the axial pressure and radial pressure
Trang 5Radial Pressure
Grease
Gland Packing
(Outside)Micro Washer
Fig.3 Comparison of measured value Fig.4 Structure of Measuring Device Fig.5 Photo of Measuring Device
METHOD OF MEASURING RADIAL PRESSURE THROUGH GREASE AND EVALUATION
OF THIS METHOD
The proposed measuring method is shown in Figure 2 A hole for sensors is made on the bonnet andthe radial pressure of the packing is measured through grease The sensor is a Miniature ThreadedPressure Sensor made by ENTORAN
The packing's exact radial pressure value may not be measured if the packing is deformed by the holefor the grease Thus the accuracy of the measurement of the radial pressure through grease wasconfirmed experimentally A stainless ring of 14.5mm in height and 13mm in thickness that can bestuffed with two pieces of packing was produced Using this ring, the packing's radial pressures asmeasured by the pressure sensor through grease and by a strain gauge were compared
The pressure inside the ring, i.e., the average value of the packing's radial pressure Pr is expressed using the circumferential strain outside the ring ee [i]
The change of radial pressure when pressed by 2, 3, and 4 N-m using a torque wrench every 15minutes is shown in Figure 3 From the graph, it is seen that the value of radial pressures measuredthrough grease and calculated from the strain outside the ring are mostly in agreement Therefore, theradial pressure value measured through grease can be stated accurately not only qualitatively butquantitatively
Trang 6Ch76-I044963.fm Page 379 Tuesday, August 1, 2006 8:30 PM Ch76-I044963.fm Page 379 Tuesday, August 1, 2006 8:30 PM
379
MEASUREMENT OF THE RADIAL PRESSURE DISTRIBUTION IN THE SYSTEM
The distribution of the radial pressure of the piled-up packings in the system was measured under theconditions described in previous section The experimental system is shown in Figures 4 and 5 Thepackings are placed firmly in a stuffing box and the stem can be operated by a control device The holefor the placement of the sensors was made at the side portion of the stuffing box, and radial pressurewas measured by the measuring method described earlier Axial pressures on both sides were measuredusing the load cell
The result is shown in Figure 6 and 7 Ten graphite packings whose inside radius, outside radius, andheight are 5mm, 11,5mm, and 6.5mm, respectively, were piled up The radial pressures are referred A,
B, C, D, and E sequentially from the undersurface side Axial pressure on the side for tightly bindingthe packings is referred to as the upper pressure and the pressure on the fixed end is referred to as thebottom pressure The stem was rotated at 0.2Hz
CONCLUSIONS
This study evaluated a method for measuring the radial pressure of gland packing of an automaticcontrol valve for the purpose of diagnosing failure, and the distribution of the radial pressure in thesystem was measured using the newly developed method
The following results were obtained:
(1) Compressibility of grease does not influence the measurement of radial pressure through grease.(2) The radial pressure measured through grease is in general agreement with the value calculatedbased on the circumferential strain outside the stainless ring stuffed with two pieces of packing(3) The hydrostatic thrust bearing model can be applied to the outflow of grease, and this outflow'sinfluence on the measurement can be disregarded by providing a margin of about one piece ofpacking on both sides of the sensor hole
Trang 8The aim of this paper is the development of a micro needle made of biodegradable polymer (PolyLactic Acid, called as PLA) This device is applicable to a blood test system for diabetics The needlehaving the size of 1000 j^m length, 64 ^m height and 127 j^m width has been developed.Biodegradable polymer naturally degrades itself in tissues; therefore this material is safe for humanbody and environment To achieve the purpose of this study, we have focused on the establishment ofwet chemical etching process and micro-molding processes for PLA The resistively of PLA againstcommon chemical etching solutions and dry etching gases is also investigated
of diabetics (including the suspect of it) is about 10% in Japan To make matters worse, the generation
of diabetics is becoming younger They have to collect their blood for the glucose level measurement
at least twice a day, which is indispensable for health monitoring In this blood collection, they feelpain and fear when needle is pricked A commercial needle is usually straight and made of metal atpresent state The new type needle decreasing pain and fear is strongly desired now On the contrary, ahuman being has almost no pain and no fear when he is bitten by a mosquito The diameter ofmosquito's needle is about 30 um, which is small enough to circumvent the pain spots of a human skin.And also, mosquito's needle has a jagged shape and it can easily cut out skins with vibration A jaggedshaped needle has a merit of decreasing the contact area between the surface of the needle and thedermis of a skin during insertion [1] Therefore it is considered that the needle can be easily insertedinto a skin with vibration Several micro needles have been reported for biomedical applications, such
as drug delivery [2], measurement of cortical biopotentials, etc However, it is difficult to form athree-dimensional jagged needle shape by micromachining technique The aim ofthis paper is to develop such shaped micro needle made of biodegradable polymer, which is assumed
Trang 9Figure 1: Schematic view of the device
to be used by the diabetics in their blood tests Figure 1 shows the schematic view of this needle In
micromachining technology, silicon is usually used as a structural material Silicon is inert material to
a human organism; however, silicon is not safe for use in medical treatment If a needle is broken andpieces of it are left in organism, they may cause a fatal problem Hence, biodegradable polymer is used
in this study This material is safe for human, since it naturally degrades itself in organism according tohydrolysis process, and is finally dissolved to harmless materials of water and carbon dioxide [3, 4].There are small numbers of report about micro-molding process for biodegradable polymer [5] In thisresearch, a jagged complicated needle is fabricated by this micro-molding method As a biodegradablepolymer, Poly Lactic Acid, called as PLA, is adopted
CHEMICAL COMPATIBILITY AND MECHANICAL PROPERTIES
In advance of microfabrication, resistance of PLA against several wet etching solutions and dryetching gases, which are commonly used in micromachining process, are investigated PLA is able to
be wet-etched by TMAH solution, and it has resistance against acids such as H3PO4, HF PLA is able
to be dry-etched by O2, SFe and CF4 plasma gases, and it has resistance against CHF3 plasma gas
Table 1 shows the summary of chemical compatibility of PLA Table 2 shows the comparison of the
mechanical properties of representative polymer, Polyimide and PLA Mechanical properties of PLAare not so inferior to Polyimide, and melting point of the PLA is lower than that of Polyimide
Tablel Compatibility of chemicals with PLA Table2 Comparison of Polyimide and PLAWet etching
Diy etching
TMAH
I IF BjPO,
o
O : No reaction with PLA (compatible) ~~|
X : Reacts with PLA (not compatible)
Young's Modulus [GPa]
Tensile Strength |MPa|
Elongation [%]
Glass Transition Temp.
Melting Point [°C]
Corporation Trade name Manufacturing method
Polyimide 3 120 10 310 450 DuPont MicroSystcms, Ltd.
PIX-3476-4L Spin coat
Poly Lactic Acid 3.4 64 4.1 61 173 Shimadzu Corp Lacty500O Injection Molding
FABRICATION AND RESULTS
Fabrication Process of Micro Needle
The micro needle is fabricated as shown in Fig 2 A (lOO)-oriented silicon wafer, of which both front
and back surfaces are covered with thermally grown SiC>2, is prepared SiC>2 mask for wet etchingwhich has jagged shape of needle is patterned by using photolithography and dry etching Then silicon
is anisotropically wet-etched to prepare pyramidal holes Consequently these holes are connectedtogether and a jagged groove is formed PLA is molded into this groove and released
Trang 10(c) Etch oxide (g) Mold and etch hack PLA Table3 Parameters for evaluating jagged groove
(d) Etch silicon (h) Release
No.
1 2 3 4 5 6 7 8 9
DiaiiKler of circlefum]
70 70 70 70 90 90 90 90 90
Displacement of square[(.un]
1.5 2.0 2.5 3.0 2.5 3.0 3.5 4.0 4.5Figure 2: Fabrication process of micro-needle
Anisotropic Wet Etching
The jagged groove of the micro needle is fabricated by anisotropic wet etching of silicon (100) surface
Tn order to produce desired jagged shape, optimal value of radius of circles and their distance are
investigated Figure 3 shows the schematic view of the mask pattern Table 3 shows the parameters
for evaluating jagged groove If the displacement of circles in Fig.3 is too close, before pyramidalholes would be made, they are connected to each other and desired jagged shape cannot be formed as
shown in Fig 4 On the other hand, the displacement of circles is too apart, the pyramidal holes are not
connected easily, which consumes large process time According to these phenomena, uniform
connection of pyramidal holes is not obtained by using only KOH solution as shown in Fig 5.
Therefore, a new process of etching silicon groove is developed First KOH solution is used forpyramidal part, since etching is stopped accurately on (111) surface, second TMAH solution is used forconnecting part, since TMAH etching forms more isotropic shape compared with KOH etching The
result of etched groove using both KOH and TMAH solutions is shown in Fig 6.
Connected partNot connectedpart
(a) 60min etching by
KOH solution
(b) 90min etching byKOH solution
Figure 4: Etched groove when squares are too close
Figure 5: Jagged groove when etched
only by KOH solution
Figure 6: Jagged groove of silicon cavity for
micro needle which is etched by bothKOH and TMAH solutions
Trang 11Ch77-I044963.fm Page 384 Tuesday, August 1, 2006 9:45 PM Ch77-I044963.fm Page 384 Tuesday, August 1, 2006 9:45 PM
384
Micro-molding
Next, micro-molding of PLA is investigated The schematic diagram of compression molding to
produce the 3D micro structures of PLA micro needle is shown in Fig 7 First, PLA is re-flowed at
above its melting temperature, and is put onto a heated mold (200°C, 3 min.) Second, pressure isapplied to the silicon cavity and PLA (14 MPa, 1 min.) Third, PLA is etched back to the level of thesilicon surface by oxygen plasma gas Finally, the micro needle is taken away from the mold by using
an adhesive tape Figure 8 shows the fabricated PLA micro needle The size of needle is as follows:
length is 1000 um, height is 64 um and width is 127 ixm
ACKNOWLEDGEMENT
This work was mainly supported by JSPS (Japan Society for the Promotion of Science).KAKENHI(16310103) This work was also partially supported by MEXT (Ministry of Education, Culture, Sports,Science and Technology) KAKENHI (17656090), "High-Tech Research Center" Project for PrivateUniversities: Matching Fund Subsidy from MEXT, 2000-2004 and 2005-2009, the Kansai UniversitySpecial Research Fund, 2004 and 2005
REFERENCES
[1] Oka K., Aoyagi S., Arai Y., Isono Y., Hashiguchi G and Fujita H (2002) Fabrication of a micro
needle for a trace blood test Sensors and Actuators 97-98C, 478-485.
[2] Henry S., McAllister D V., Allen M G and Prausnitz R (1998) Micromachined needles for the
transdermal delivery of drugs Proc ofMEMS'98, 494-498.
[3] Tuji S and Ikada Y (1997) Poly Lactic Acid -for medical care • medication • environment-
KOBUNSHIKANKOUKAI, LTD., 1-76.
[4] Urakawa H and Ohara H (2000) Crystallinity and molding properties of Polylactic Acid
Technology Research Laboratory, Shimadzu Corp., 56:3*4, 163-168.
[5] Armani D K and Liu C (2000) Microfabrication Technology for Polycaprolactone, a
biodegradable polymer J.Micromech.Microeng 10, 80-84.
Trang 12M Nunobiki: K Okuda1 and S Maeda2
1 Department of Mechanical and System Engineering, University of Hyogo,
2167 Shosha, Himeji, Hyogo, JAPANDepartment of Quality Assurance, Shin Caterpillar Mitsubishi LTD
1106-4 Shimizu Uozumi, Akashi, Hyogo, JAPAN
ABSTRACT
One of the problems in reinforcement learning with real robots is to need a large number of trials Thispaper proposes a reinforcement learning that uses fuzzy ART for segmentation of state-space.Whenever fuzzy ART encounters a new situation, it generates a new category node to the state-space
We proposed generating methods of new category nodes that inherit the state-value and the policyfrom a similar node Proposed methods were estimated from simulations of a two-link manipulatorand a multi-link mobile robot It was confirmed that the proposed method was able to increase thelearning speed and reduce the size of state-space
an action from a continuous-valued action, the performances are insufficient to apply to the real robotsyet (Morimoto and Doya.) A grid-like representation of the state-space was insufficient to the
Trang 13REINFORCEMENT LEARNING WITH FUZZY-ART STATE-SPACE CONSTRUCTION
Figure 1 shows the architecture of proposed system This system based on actor-critic methods Usingthe sensed state from environment, the actor selects an action The critic evaluates the new state todetermine whether the state has gone better or worse than expected The TD-error represents thisevaluation The TD-error is used for the improvement of the policy and the update of the state-value IfTD-error is positive, it suggests the tendency to select the action should be strengthened for the future
If TD-error is negative, it suggests the tendency should be weakened We applied the fuzzy ART togeneralization of sensed states In many tasks, most states will never have been experienced exactlybefore Fuzzy ART is a kind of self-organized clustering method and it classifies unknown state intothe group of approximate states One state code used for each group
FUZZY-ART
The fuzzy ART consists of two fully connected layers The category layer contains category nodes tocategorize a given vector A weight vector Wj shows the representative pattern of category node j Thesensed data are normalized with complement cording The sensed data vector and its complementvector are input to the input layer The choice function Ti is defined as equation (1) Where c is achoice parameter, the operator is defined by (p A q) =min(p, q) and the norm | x | is the sum of itscomponents The category node i that has the maximal Ti is called a winner node The winner node isjudged to cause resonance or mismatch reset according to the equation (2) If the match function Mi isbigger than the vigilance parameter p, resonance is carried out and the weight vector Wi is updatedaccording to the equation (3) Where (3 is a learning rate parameter When mismatch reset procedureshould be done, other category node that has the next maximal T is chosen again When any categorynode is not selected, a new node is generated In normally, default values are given to the policy andthe state-value In proposed methods, inherited value from a similar node was used The efficiency wasestimated in the simulations of a two-link manipulator and a multi-link mobile robot
Fuzzy ART
i l l A'1'2 ATi A' Category L a y e r ^ ^ ^ ^
Trang 14Ch78-I044963.fm Page 387 Monday, August 7, 2006 11:30 AMCh78-I044963.fm Page 387 Monday, August 7,2006 11:30 AM
387
LEARNING FOR HAND REACHING PROBLEM OF TWO-LINK MANIPULATOR
Figure 2 shows a hand reaching problem and the results of learning A reward of 1 was given when theend effecter reached the goal in 50 steps or less Each angle of joint was limited to 0<6<2TI Eachangular velocity was limited under 0.1 rad/step We addressed a state with the x and y coordinates andthe velocity components of the end effecter The step-size parameter and the credit rate were set to0.02 and 0.9 respectively The learning methods were estimated by an attainment rate of task and thenumber of category nodes The attainment rate of task was defined as the ratio between number of thesucceeded trials and the total trails We carried out the simulation on each method five times and usedthe average value The vigilance parameter was set to 0.98 This result shows that the methods withfuzzy ART were superior to the normal method Table 1 shows the attainment rates and the number ofcategory nodes at 10,000th trial at various methods and various p The attainment rate increased sothat p became large, hi case that p was 0.98, all methods are superior to the normal method Thelearning method of inheriting the state-value was superior to other methods In a normal actor-criticmethod, we tiled the sate-space with a uniform grid and set the total number of states to 10,000 Inlearning with fuzzy ART, the number of category units increased so that p became large, but the totalnumber of states at 10,000th trials was 202.2 or less
100mm
Normal method 1'uzzy ARI without inheritam (,« =0.98)
stato"valuc O =0.98) Fuzzy ART with inheritance of policy ( p =0.98) FuzzvART willi inheritance of
—-^^Vigilance criterion P
S t a t e - s p a c e ^ ^ ^ ^ ^ ^ ^ ^ segmentation method ^ ^ - ^ ^ ^ Fuzzy ART without inheritance Fuzzy ART with inheritance ol'state value
Fuzzy ARF with inheritance ofpolicv
Fuzzy ART with inheritance ofpolicv and state value Normal method
Attainment rate of task 0.95
76.8%
75.5%
68.2%
0.97 83.0%
88.0%
86.0%
80.1%
0.98 94.4%
0.96 85.4 79.6 78.4 85.4
0.97 121.2 115.0 108.8 124.6
0.98 191.0 182.8 191.6 202.2 10000
LEARNING FOR MOVEMENT EXPERIMENTS OF A MULTI-LINKED MOBILE ROBOT
We applied proposed method to multi-liked mobile robots The simple model was used It consisted offive servomotor modules that were connected serially The cyclometer was installed to measure themoved distance from the initial position The moved distance by 20 steps was given as the reward But
a negative reward was given when the robot had moved backward Each trial began with same posture
in which all joints are stretched 100 trials were carried out We addressed a state with angles of fourservomotors The step-size parameter and the credit rate were set to 0.02 and 0.9 respectively Thevigilance parameter was set to 0.98 The learning methods were estimated by an attainment rate of task
Trang 15Ch78-I044963.fm Page 388 Monday, August 7, 2006 11:30 AM Ch78-I044963.fm Page 388 Monday, August 7,2006 11:30 AM
388
and the total moved distance by 100 trials Table 2 shows the attainment rate of task at 100th trial and
the total moved distance The attainment rates were low because the number of trials was a little The
learning methods with fuzzy ART were superior to the normal actor-critic learning method
Furthermore, the learning speed had quickened most and the total moved distance was largest in using
inheritance of the state-value The efficiency of the proposed method was confirmed in the experiment
of multi-link mobile robot though the number of trials was a little
TABLE2 PERFORMANCE OF EACH LEARNING METHOD FOR A MOBILE ROBOT
Module2 Module4 Modulel / Module3 / Module5 Cvclomctcr
Total moved distance Attainment rate of taskat 100 th trials
Normal method
429 mm
2 9 %
Fuzzy ART without inheritance
485 mm
3 4 %
Fuzzy ART with inheritance
578 mm
3 6 %
CONCLUSION
We proposed a reinforcement learning method that used fuzzy ART for segmentation of state-space
And we proposed a generating method of a new category node that inherited the state-value of the
similar node The efficiency of proposed method was estimated in the simulations of hand reaching
problems and the movement experiments The learning efficiency was improved more by inheriting
the sate-value in the fuzzy ART The learning speed of proposed method is about 20 times the speed of
normal actor-critic method in the hand reaching problems The size of state-space was decreased very
much in proposed method The efficiency of proposed method was confirmed in the experiments of
multi-link mobile robot Thus, it was confirmed that the proposed method was able to apply to the
learning with the real robots
References
Y Takita, M Nunobiki, et al (1999) An investigation of climbing up stairs for inchworm robot,
Proceedings ofTITech COE/Super Mechano-Systems Workshop'99, 133-138
M Nunobiki, T Takita, et al (1999) Study on the gaits of an inchworm robot through a narrow
path Advanced Robotics, 13: 3, 329-330
M Nunobiki, et al (2001) An investigation of mobility of inchworm-type mobile robot,
Proceedings of the France-Japan Congress of Mechatronics, 113-118
M Nunobiki, et al (2004) Returning motion from the state of falling sideways for articulated
mobile robot, JSME International Journal, Series C, 47:1, 225-232.
R.S Suttun and A.G Barto (1998) Reinforcement Learning; An Introduction , The MIT Press
A.G Barto, R.S Suttun and C.W Anderson (1983) Neuronlike adaptive elements that can solve
difficult learning control problem, IEEE Trans Syst., Man, Cyber., SMC-13:5, 834-847
J Morimoto and K Doya (2001) Acquisition of stand-up behavior by a real robot using
hierarchical reinforcement learning Robotics and Autonomous Systems, 36, 37-51
J Morimoto and K Doya (1998) Reinforcement learning of dynamic motor sequence: Learning
to stand up, Proc of IEEE/RSJ Int Conf On Intelligent Robots and System, 3, 1721-1726
G.A Carpenter, S Grossberg and D.B Rosen (1991) Fuzzy ART: Fast stable learning and
categorization of analog patterns by an adaptive resonance system Neural networks, 4, 759-711