Intelligent Control of Electric Scooters

D.T. Lee∗ C. H. Wu
S.J. Shiah Department of Electrical and Computer Engineering
C.M. Lee Northwestern University
Institute of Information Science Evanston, IL 60208
Academia Sinica email: [email protected]
Taipei, Taiwan
email:{dtlee,bioman,cmlee}@iis.sinica.edu.tw

ABSTRACT to commute to and from work. However, the amount of

Most of the present day electric scooters are equipped with exhaust and noxious particles produced by the scooter is
a voltage-driven DC motor powered by four 12-volt lead- quite significant that the scooter is considered a moving air
acid batteries and a hand-lever accelerator operated by the and noise pollutant and is affecting the air quality and living
rider to control their speed. Because of the nonlinear bat- environment seriously[1].
tery discharge characteristics and different driving behav-
iors of riders, it is not easy to tell how much electric power In an effort to reduce air pollution, Taiwan govern-
remaining in the battery and how far the electric scooter can ment has implemented a subsidiary program to encour-
travel before the battery has to be re-charged. As a result, age research and development of battery powered electric
the reliability of the electric scooter is lacking. scooters since 1997. However, commercially available
To tackle this problem and to enhance the capabilities electric scooters to date have some drawbacks, including
of present electric scooters, we propose an intelligent con- high cost, long battery recharging time, relatively short
trol system that not only can control the speed of the elec- traveling distance for each re-charge, and inadequate feed-
tric scooter, but also can provide information about resid- back information to the user with respect to the residual
ual electric power in the battery system by monitoring its battery power. The lack of a more reliable or accurate
power consumption. power prediction and management mechanism often results
This system consists of both motor driver control and in situations that the riders unwittingly run out of battery
energy management subsystems. The driver control sub- power before they reach their destinations or a facility to re-
system is implemented as a closed-loop speed control sys- charge the battery. This uncertainty as to when the battery
tem by using a muscle-like control law with excellent com- power will run out, could be rather troublesome and there-
pliant property. The energy management subsystem is fore hinder the sale of electric scooters. The crux in get-
implemented by learning modules based on fuzzy neural ting a good estimate of the present battery state-of-charge
networks and cerebellar model articulation controller net- (SOC) lies in the nonlinear battery discharge characteristics
works, which can estimate and predict nonlinear charac- which vary with the road condition and the driving behav-
teristics of the power consumption of batteries and electric ior of the rider.
scooter dynamics. With this battery power monitoring sub- To tackle the reliability problem of present electric
system the rider will be provided information regarding an scooters, we propose an intelligent system which consists
estimated traveling distance at a given speed, and the maxi- of a motor driver control subsystem and an energy man-
mum allowable speed to guarantee safety arrival at the des- agement subsystem for controlling and monitoring power
tination with the residual battery capacity. consumption of electric scooters. The motor driver con-
Experimental results show that the performance of trol subsystem (DCS) provides a closed-loop speed control
electric scooters can be improved substantially. and is implemented by using a muscle-like control law[2].
Its excellent compliant and short-range enhancement prop-
KEY WORDS erty can give the rider the needed feeling of acceleration
Electric Scooter, Intelligent Control, Muscle-like Compli- and maneuverability. In contrast with the open-loop con-
ant Control, Fuzzy Neural Network, Cerebellar Model Ar- trol adopted by the present electric scooters, the closed-
ticulation Controller loop control enables the rider to fully control the speed of
the electric scooter. In so doing, the rider can control the
consumption of electric power via the proposed subsystem.
1 Introduction The energy management subsystem (EMS) will provide the
rider with the critical information including estimated al-
There are more than ten million gasoline powered scooters lowable traveling distance, safe speed, and remaining bat-
or motorcycles being used in Taiwan every day. They have tery capacity etc., such that the rider can control the speed
become a very common transportation means for people of the scooter within a proper range via DCS and en-
sure that the scooter can arrive at the destination safely. 2.1 Drive Control Subsystem(DCS)
We use learning modules which consist of fuzzy neural
networks (FNN)[3] and cerebellar model articulation con- Inspired by the compliant capabilities of the biological
trollers (CMAC) [4, 5] to estimate nonlinear characteristics limb, an active damping control based on a muscle-like
of the power consumption of batteries and electric scooter compliant control is adopted and implemented in the ker-
dynamics. Online learning about power consumption is a nel motor-control subsystem for controlling an electric
unique feature of our EMS. It is incorporated into this sub- scooter. The proposed muscle-like control model was fit-
system to make it more adaptable to different electric scoot- ted from the responses of voluntary and involuntary limb
ers and driving behaviors. movements[6, 7, 8, 9]. Because of its compliant adaptabil-
ity, the controller is very well suited for man-machine in-
terface control[2], such as the hand accelerator in an elec-
2 System Implementation tric motorcycle or the foot accelerator in an electric vehi-
cle. Because of its unique nonlinear damping property, the
The conceptual organization of the proposed intelligent muscle-like control enables an electric scooter to adapt to
control system is shown in Fig.1. As stated earlier, this sys- varying loads and sudden impacting forces. The main dif-
tem consists of two subsystems, and they are motor driver ference between the proposed controller and the conven-
tional controller currently used by commercially available
electric scooters is that the former is a closed-loop, com-



   

pliant controller while the latter is an open-loop controller.


Furthermore, it can provide the rider with the feeling of
  
  great acceleration just like the open-loop control, and with





cruise control capability by allowing a desirable speed to
422,3/

be set and maintained by DCS. The cruise control function

   


may be disabled manually or automatically when the brake
 
of the electric scooter is applied. Thus, the current or elec-
tric energy consumed by manual acceleration and decelera-

 tion operations can be reduced, and the life of the battery or


traveling distance can be prolonged. The muscle-like com-
pliant control model consists of two major parts: spindle-
like mechanism emulating the reflex property of biologi-
Figure 1. An illustration of the proposed intelligent control cal muscular system for absorbing impacting forces, and
system muscle-stiffness mechanism emulating muscle stiffness for
tracking various movements. The control blocks of the
DCS are shown in Fig.2. Both inputs of DSC are the veloc-

control subsystem (DCS) and energy management sub-

system (EMS). The rider sends speed commands to DCS
 
  
 


through a user interface which could be a lever, button, or a   
.P
touch pad, and the commands will be converted into proper
 
;2
 



  
signals obtained from the muscle-like control law to drive ∫
9GHVLUHG  2:8   


/ 2:8
the electric scooter according to the feedback velocity sig- [
nals. The signal can drive the scooter to move as fast as  
!
it could to the extent as if it were operated under an open-  
7
2:8 
 ∫
loop control, so the rider does not have to change driving
'GHOD\ "

behavior. The muscle-like control also possesses a capabil- 
'
ity to filter out noisy or high-frequency disturbance input
signals. It lets the scooter react to speed commands even
the road condition changes constantly. On the other hand,
the energy management system (EMS) monitors the power Figure 2. An illustration of the Drive Control Subsystem
consumption of batteries and by learning the battery dis-
charge characteristics, computes the maximum travel dis-
tance and safe speed, and provides this important informa-
tion to the rider through a user interface. When the rider ity command Vd and the feedback velocity signal V . The
gets the important information, he/she will be able to make motor command C mus used in muscle-like model is de-
adjustments to the speed of the scooter by using DCS. fined as


The details of these two subsystems will be discussed in the
following sections. C mus = (Vdelay − Vdesire )dt (1)
The reflex signal of the spindle-like model, r mus , scaled load voltage method[14, 15], ampere hour accumulation
through a reflex gain coefficient, H, is combined with the method[16, 17, 18], and internal resistance method[19].
motor command to produce a reflex-induced command for The load voltage method is suitable for constant load cur-
muscle force. The linear feedback gain L represents the rent applications. For electric soccters, it cannot provide
effect of muscle length-tension which will vary with load precise estimations, because the load current will vary a lot
position. Any change in muscle length will produce a mus- with the road condition, and the load and speed change of
cle force through muscle stiffness K m . The muscle force the scooter. The ampere hour accumulation method is used
used in the present implementation of DCS is defined as to accumulate the discharge current, and estimate the SOC
the DC motor control voltage v m . It can also be defined as according to this accumulated value and pre-recorded data
the DC motor control current, depending on the type of DC describing the relations between battery discharge current,
motor driver used. Thus, the control law for generating v m voltage, and capacity. The pre-recorded data is not valid in
can be expressed as every discharge condition, so methods making use of mod-
els of fixed parameters to estimate SOC will suffer from
vm = Km (Lxmus + LP F (Hrmus − C mus )) (2) loss of precision when the discharge conditions are chang-
ing over time. The internal resistance method needs to mea-
where LP F (·) represents a low pass filter. The spindle- sure the frequency response of the battery to determine its
like model with the nonlinear, fractional, damping effect SOC. Because it needs extra electric circuits and function
multiplied by the short-range elasticity enhancement cab generators, it is difficult to implement.
be expressed as In our EMS, we use four learning modules, as shown
1 in Fig.3. These four learning modules are implemented
Bp ẋpn (|xp | − xp0 ) = Kr (xmus − xp ) with FNNs and CMAC networks, as shown in Fig. 4.
= rmus (3)

where xp , ẋp , and xp0 are the internal position, velocity,
   
  


and bias position of the spindle model, n is an odd integer
(n=5 in our case), B p is a damping coefficient, and K r is
,N
the reflex stiffness. The muscle equilibrium position x mus 
    
    62+ N

used in DCS is expressed as ' 07747 

'N 
    LQLWLDO

'0


   

xmus = (Vdelay − V )dt (4)

ηN
 62& N
 UHPDLQ


N  
where Vdelay and V are scooter velocities at the kth and  GLV
N

(k+1)st sampling times, respectively. According to Eq.(3),

any reflex signal will induce a corresponding force to re-
spond to a change of scooter velocity. This nonlinear
damping can deal with disturbance variations within a large Figure 3. An illustration of the Energy Management Sub-
range, and reduce the effect of parameter variation of a non- system
linear system.

2.2 Energy Management Subsystem(EMS)

The proposed energy management subsystem possesses 62+ N

62+ N

functions of battery SOC prediction, battery state of health
)11 :HLJKWV
)11 :HLJKWV

(SOH) prediction, maxmium travel distance estimation, ,

η ,N '0
N
 

N

and safe speed estimation. The performance of this sub-  UHPDLQ

LQLWLDO
-
-
N


  GLV
N
' HUURU 9N

system depends on whether we can find a suitable method D

E

or mechanism to model the nonlinear battery charge and ' 07747


discharge dynamics. Learning controllers which are bio- , N 62+ N
)11 :HLJKWV

 62+ N
logically inspired and intended to model human experience '
 UHPDLQ
N
NGLV N

[10, 11, 12] are an attractive alternative to dealing with non- 
LQLWLDO


  UHPDLQ
 &+$
linear systems of incomplete models or inaccurate model N

-  '&+
F G

parameters. We therefore decide to use FNNs and CMAC

networks [13] as the core of this subsystem.
In general, the battery capacity is a nonlinear func- Figure 4. (a) Learning Module I (b) Learning Module II (c)
tion of discharge current, temperature, depth of discharge, Learning Module III (d) Learning Module IV
and recharging times. At present, there are many meth-
ods for estimating the SOC, and they can be classified as
 dis
In Fig.3, system inputs are the battery voltage V k and where Ck∗ is the relative released capacity when the
the discharge current I k . The Learning Module I (LM I) scooter is turned off.
represents the nonlinear mapping between the discharge The structure of combination of an FNN and a CMAC
current, relative remaining capacity C kremain , SOH, and network is used to implement learning modules shown in
battery discharge efficiency η Ik . This module will send Fig.4(a),(b), and(d). In these modules, the CMAC network
discharge efficiency η Ik to the Coulomb Counting Mod- will send a set of weights to the corresponding FNN ac-
ule (CCM), and the relative battery capacities at kth and cording to its input. Then the FNN starts sending output
(k − 1)st sampling times respectively can be expressed as signals. The learning process of the FNN begins and the
ηIk dis updated weights of FNN are used as training patterns for
CIdis = C + Ik ∆T (5) the CMAC. So every learning process of learning mod-
k
ηIk−1 Ik−1
ules includes FNN learning, CMAC learning, and their
CIdis = I1 ∆T (6) interactive learning. This kind of learning structure can
1
enlarge the learning space coverage on complex dynamic
where ∆T is the sampling time, and and C Idis
are k
CIdis
k−1 systems[13] so as to increase the speed of convergence of
capacities released from the battery at kth and (k − 1)st learning modules used in the battery management applica-
sampling times respectively . Thus, the SOC at the kth tions.
sampling time can be expressed as By using learning modules, the EMS can update pa-
  rameters by itself with the change of dynamics of the bat-
CIdis
SOCk (%) = 100 1 − k
(7) tery. This design avoids the unprecision resulted from
C0 the direct estimation of battery dynamics, and can govern
the battery discharge characteristics much better than the
where C0 is the battery rated capacity. The CCM imple-
method of linearly modifying SOC according to a fixed pa-
mented by Eq.(5)-(7) is also the feedback signal provider
rameter model.
to LM I. The Learning Module II (LM II) represents the
nonlinear mapping between I k , Ckdis , SOH, and Vk . The
difference Verror will be used to update weights of FNNs 3 Experiment
and CMACs in this module. The learning rule for LM II is
designed as Two parts of this experiment are described in the following
1
E = (Vk − Ve )2 (8) sections. One is for DCS, and another is for EMS.
2
So the amount of updating weights of FNN can be ex-
pressed as 3.1 Testing for DCS
∂E We used a commercially available electric scooter to con-
∆WF N N = (9)
∂WF N N duct our experiment and road tests. A notebook computer
∂Ve was used to implement the intelligent control system. All
= (Ve − Vk ) (10) the software code was written in C++ by using Borland
∂WF N N
(11) C++ Builder 5.0, and the control signal calculated by the
proposed control law was sent to the motor via the connec-
Meanwhile, the learning rule for LM I can be designed as tion circuit and a DAQ card. The scooter for the experi-
∆WF N N =
∂E
(12)
ment is shown in Fig.5(a). A rider performed the road test
    
∂WF N N

∂E ∂Ve ∂Ck dis ∂ηk

= (13)
∂Ve ∂Ck dis ∂ηk ∂WF N N
 C dis
 
∂Ve k−1 ∂ηk
= (Ve − Vk ) (14)
∂Ck dis ηk−1 ∂WF N N

The Learning Module III represents the nonlinear mapping

between Ik , Ckremain , and SOH. Its output is the estimated
SOH, and the desired SOH at the kth sampling time is de-
signed as  
ηk
SOHk (%) = 100
d
(15)
ηk−1
The Learning Module IV represents the mapping between D E

Vk , Ckremain , SOH, and initial capacity C initial . This mod-

ule is used to estimate the remaining capacity when the Figure 5. (a) The electric scooter for the experiment (b)
rider turns on the scooter. The desired C initial is designed Road test
as
Cinitial = C0 − Ck∗ dis
(16)
according to conditions of our setup. We tested the acceler- speed at 30 km/hr. Then this process was repeated, with
ation property first. The rider would accelerate the scooter another passenger weighing 71 kg traveling together. In
from stop to a speed at 40 km/hr with open-loop control, Fig.8, the speed rising time is shown to have a delay of 2.1
and this process was repeated by using the muscle-like con- second when the total payload is 136 kg. This case demon-
trol. The results obtained are shown in Fig.6. In Fig.6(a), strates that the muscle-like controller is insensitive to the
the difference of speed rising times between the open-loop load of the scooter.
control and muscle-like control is about 1.5 seconds, and
the rider feels no significant difference between these two 

control modes. Fig.6(b) shows that the muscle-like con- 
  

trol saves about 30 % electric energy. In the second step, 



þ Âû 






  

    


NPKU




     
 






                 


 

   
     
   
     

D


 

 
  
 
 


,25
80





                

 Figure 8. Results of Speed Response to Varying Loads
 

Figure 6. (a)Scooter speed under two control modes

(b)Energy consumption difference under two control
3.2 Testing for EMS
modes
According to the motorcycle driving patterns estiblished by
the Bureau of Standards of Ministry of Economic Affairs
the rider drove the scooter to slide down from a slope, start- of Taiwan, a power consumption curve can be estimated,
ing at a point marked with x in Fig.7. When it entered a flat, and it is shown in Fig.9. The battery was discharged ac-
its speed was reduced to 10 km/hr. Then, the rider set the
desired speed at 20 Km/hr at a point marked with y in Fig.7.
At the point marked z in Fig.7, the scooter climbed another ÄÃÃÃÁÃ

slope. The current flowing into the motor increased, which ËÃÃÁÃ
reflects that controller intends to maintain the speed of the
ÉÃÃÁÃ
ã
ø»êô¼

scooter. This demonstrates the muscle-like controller’s ca-

ÇÃÃÁÃ
pability of adapting to road conditions automatically.
ÅÃÃÁÃ

ÃÁÃ
ÉÃ

 


[
ÈÃ ÃÁÃ ÅÃÁÃ ÇÃÁÃ ÉÃÁÃ ËÃÁÃ ÄÃÃÁÃ ÄÅÃÁÃ
ÇÃ \
ÆÃ çü ø»øö¼
]

ÅÃ
ÄÃ
Ã
Ä ÊÈ ÄÇÌ ÅÅÆ ÅÌÊ ÆÊÄ ÇÇÈ ÈÄÌ ÈÌÆ ÉÉÊ ÊÇÄ ËÄÈ ËËÌ

 
D

ÆÃ Figure 9. Power Consumption Curve

 

ÅÈ
ÅÃ
ÄÈ
ÄÃ
È
Ã
Ä ÊÉ ÄÈÄ ÅÅÉ ÆÃÄ ÆÊÉ ÇÈÄ ÈÅÉ ÉÃÄ ÉÊÉ ÊÈÄ ËÅÉ ÌÃÄ

 
cording to this curve repeatedly until the battery voltage
E
dropped under the cutoff voltage, 10.25V. We then charged
this battery to its stantard voltage 13.2V. The discharge and
Figure 7. Change of Road Condition Test charge process is called a testing cycle. We repeated the
testing cycle, and checked how the EMS performed. The
battery under test is a sealed lead-acid battery with a rating
of 12V open-circuit voltage and 50Ah capacity, manufac-
The final test was first performed by a rider with tured by Long Battery Co. The programmable system used
weight 65 kg. He accelerated the scooter from stop to a to perform testing cycles is manufactured by DIGATRON
Co., and the charge and discharge currents provided range
between 0A and 200A. The EMS was implemented in the ÄÅÃ

same way as the DCS. Initial weights of all learning mod- ÄÃÃ

ules were trained by using 98 data sets of constant current ËÃ

æâÖ»¸¼
discharge process. Every data set included a time history ÉÃ
5HODWLYH
of battery voltage, discharge current, discharge efficiency, ÇÃ

and capacity. ÅÃ

Fig. 10(a) shows the relative SOC predicted by our Ã

proposed EMS and SOC which computed by using ampere Ã È Äà ÄÈ

hour accumulation method at the 1st testing cycle. At the çü ø» ü
¼

ÄÅÃ
ÄÃÃ
ËÃ
æâÖ»¸

ÉÃ
ÇÃ
Figure 11. The battery SOC predicted by EMS and ampere
ÅÃ
Ã
hour accumulation method at the 20th testing cycle
Ã Å Ç É Ë Äà ÄÅ ÄÇ
çü ø» ü
¼
(a) (b)
D

ÄÉ ÈÃ

ÄÇ
 
 
 ÇÃ

      


ÆÃ
ÕôÁ éÿôúø»é¼

ÄÅ
ÅÃ

ÄÃ
 
 ÄÃ

Ã
Ë
Ã Å Ç É Ë Äà ÄÅ ÄÇ
Ã Å Ç É Ë Äà ÄÅ ÄÇ
çü ø» ü
¼
ÄÇ
6HQVHG ÄÇÃ

ÄÅ ÄÅÃ
çü ø» ü
¼

×üöûôúø Öø
»ô ¼
ÄÃ ÄÃÃ

Õôø éÿôúø»é¼
E Ë ËÃ
F
É
(VWLPDWHG ÉÃ

Ç ÇÃ

Å ÅÃ

à Ã

Figure 10. (a) The battery SOC predicted by EMS and à ÃÁÈ Ä ÄÁÈ

çü ø» ü
¼
Å ÅÁÈ Æ Ã ÃÁÈ Ä ÄÁÈ

çü ø» ü
¼
Å ÅÁÈ Æ

ampere hour accumulation method at the 1st testing cy- D E

cle (b)The sensed battery voltage and voltage predicted by ÄÅÃ

ÄÃÃ
EMS (c)The relative battery capacity predicted by EMS ËÃ
æâÖ»¸

ÉÃ
ÇÃ
ÅÃ
Ã
à ÃÁÈ Ä ÄÁÈ Å ÅÁÈ Æ
çü ø» ü
¼
first testing cycle, all learning modules started to learn the F

battery dynamics, and the learning rate is set to be 0.005

for all modules. Fig. 10(c) shows that the battery cutoff Figure 12. (a) The sensed and predicted battery voltage (b)
voltage is reached after 10 discharge patterns are executed. The discharge current pattern (c) The relative SOC
Fig. 10(a) and (b) shows that the SOC prediction error of
EMS is 10 % at the end of testing, and we can see that the
SOC predicted by using ampere hour accumulation method
is 80 %. Therefore, the relative SOC shown in Fig.10(a) is
used for checking whether the scooter speed is suitable to
drive its DC motor. The estimated discharge efficiency de-
termines the precision of relative SOC. Fig.11 shows the control technique to provide a closed-loop control of the
SOC predicted at the 20th testing cycle. The EMS predic- speed the electric scooter. It provides the rider with cruise-
tion error is reduced to 7%, but the error for using ampere control capability. It also contains a battery energy man-
hour accumulation method is still 80%. agement subsystem to monitor the power consumption of
Fig.12(a) shows the sensed and predicted battery voltage. the battery. The energy management subsystem provides
When we discharged the battery according to the discharge to the rider important information regarding the residual
current pattern shown in Fig.12(b), we obtained the relative battery power of the electric scooter in terms of traveling
SOC estimation result shown in Fig.12(c). We can see that distance, and maximum allowable traveling speed to guar-
the SOC curve is almost a monotonically decreasing curve, antee safety arrival at the destination.
and this property suggests that a reliable battery capacity The proposed intelligent control system has greatly
information can be provided to the rider. improved the performance of present electric scooters. Ad-
ditional safety features, such as range sensors, can also be
4 Conclusion added to detect obstacles in the pathway of the electric
scooter to avoid collision. For instance, the speed of the
We have proposed an integrated intelligent control system electric scooter can be reduced automatically by the sys-
for improving the performance of present electric scoot- tem and an alarm may be turned on to alert the rider when
ers. This system utilizes a non-linear damping compliant such obstacles are detected.
References [13] K.Y. Young and S.J. Shiah, An approach to enlarge
learning space coverage for robot learning control,
[1] M.J. Schwarz and J.W. Shiller, Clean air act amend- IEEE Trans. Fuzzy Syst., vol. 5, no. 4, pp. 511–522,
ments of 1990, impact on motor vehicle and the en- 1997.
vironment, in Proceedings of Air & Waste Manage-
ment Association 84th Meeting Exhibition, pp. 16–21, [14] P. Finger Eugene and N. Y. Brewsler, Battery of
1991. charge metering method and apparatus, U.S. Patent, ,
no. 4560937, 1985.
[2] C.H. Wu, K. S. Hwang, and S. Lehman, Analysis
and implementation of a neuromuscular-like control [15] Simmonds et al., Device for indicating the resid-
for robotic compliance, IEEE Transactions on Con- ual capacity of secondary cells, U.S. Patent, , no.
trol Systems Technology, vol. 5, no. 6, pp. 586–597, 5479085, 1995.
1997.
[16] Kopmann, Method of and apparatus for monitoring
[3] C.T. Lin and C.S.G. Lee, Reinforcement struc- the state of a rechargeable battery, U.S. Patent, , no.
ture/parameter learning for neural-network-based 5518835, 1996.
fuzzy logic control systems, IEEE Trans. Fuzzy Syst.,
[17] Kozaki, Remaining battery capacity meter and
vol. 2, no. 1, pp. 46–63, 1994.
method for computing remaining capacity, U.S.
[4] J.S. Albus, A new approach to manipulator control: Patent, , no. 5691078, 1997.
the cerebellar model articulation controller (CMAC), [18] A.S. Clegg and C. England, Battery monitor which
ASME J. Dynamic Syst., Measurement, Contr., vol.
indicates remaining capacity by continuously moni-
97, no. 3, pp. 220–227, 1975. toring instantaneous power consumption relative by
[5] J.S. Albus, Data storage in the cerebellar model artic- expected hyperbolic discharge rates, U.S. Patent, ,
ulation controller (CMAC), ASME J. Dynamic Syst., no. 5394089, 1997.
Measurement, Contr., vol. 97, no. 3, pp. 228–233, [19] F. Huet, A review of impedance measurements for
1975. determination of the state of charge or state of health
[6] C.H. Wu, K. Y. Young, and J. C. Houk, A of secondary batteries, J. Power Sources, vol. 70, pp.
neuromuscular-like model for robotic compliance 59–69, 1998.
control, in Proc. of 1990 IEEE Int. Conf. On Robotics
and Automation, 1990, pp. 1885–1890.

[7] C.H. Wu, K. Y. Young, K. S. Hwang, and S. Lehman,

Analysis of voluntary movements for robotic control,
IEEE Control Systems Magazine, vol. 2, no. 1, pp. 8–
14, 1992.

[8] C.H. Wu and S. L. Chang, Implementation of a

neuromuscular-like control for compliance on a puma
560 robot, in Proceedings of the 34th IEEE 1995
International Conference on Decision and Control,
1995, pp. 1597–1602.

[9] C.H. Wu, S. L. Chang, and D. T. Lee, A study of

neuromuscular-like control for rehabilitation robots,
in Proceedings of the 1996 IEEE International Con-
ference on Robotics and Automation, 1996, pp. 1178–
1183.

[10] C. Alippi, A. Ferrero, and V. Piuri, Artificial intelli-

gence for instruments and measurement applications,
IEEE Instrumentation & Measurement Magazine, vol.
12, pp. 9–17, 1998.

[11] P.D. Wasserman, Neural Computing–Theory and

Practice, Van Nostrand Reinhold, NewYork, 1989.

[12] H.J. Zimmermann, Fuzzy Set Theory and Its Applica-

tions, MA: Kluwer, Norwell, 3rd edition, 1996.