A Review of Exoskeleton-type Systems and Their Key Technologies 2008

1599

A review of exoskeleton-type systemsand their key technologiesC-JYang, J-F Zhang,Y Chen,Y-MDong, andY ZhangState Key Laboratory of Fluid Power Transmission and Control, Zhejiang University, Hangzhou 310027,Peoples Republic of China

The manuscript was received on 8 November 2007 and was accepted after revision for publication on 18 March 2008.

DOI: 10.1243/09544062JMES936

Abstract: The exoskeleton-type system is a brand new type of manmachine intelligent system.It fully combines human intelligence and machine power so that machine intelligence andhuman operators power are both enhanced. Therefore, it achieves a high-level performancethat neither could separately. This paper describes the basic exoskeleton concepts from biologi-cal system to manmachine intelligent systems. It is followed by an overview of the developmenthistory of exoskeleton-type systems and their two main applications in teleoperation and humanpower augmentation. Besides the key technologies in exoskeleton-type systems, the researchis presented from several viewpoints of the biomechanical design, system structure modelling,cooperation and function allocation, control strategy, and safety evaluation.

Keywords: exoskeleton-type system, manmachine intelligent system, review, key technology

1 INTRODUCTION

The exoskeleton-type system is a kind of manmachine system centered by human. It is alwaysdesigned as an external mechanical structure whosejoints correspond to those of human body or limbs.It combines the human intelligence and the machinepower so that it enhances the intelligence of themachine and the power of the human operator. As aresult, the human operator can achieve what he is notcapable of by himself [13].

The research of exoskeleton-type systems boomedin the late 20th century, when researchers from theUnited States, Japan, Germany, and other countriesintroduced many novel concepts of manmachinesystems. Then the exoskeleton-type systems havebeen developed rapidly accompanied with greatachievements in mechanical and electronic engineer-ing, automation technology, biological, and materialscience. Due to the bottleneck of the artificial intel-ligence and the limitation of the automatic robotics,the exoskeleton-type system through its special

Corresponding author: State Key Laboratory of Fluid PowerTransmission and Control, Zhejiang University, Hangzhou 310027,

Peoples Republic of China. email: [email protected]

advantages has become a highlight in the field ofmechatronics and robotics. Its exciting applicationshave covered the robotic bilateral teleoperation withforce feedback, virtual reality (VR), entertainment [4]as well as the power amplifier and physical rehabilita-tion. This paper reviews its development history andpresents the current advancements of exoskeleton-type systems. Also discussed are biological and arti-ficial concepts, important research topics, and espe-cially the key technologies of the exoskeleton-typesystems.

2 THE BASIC CONCEPTOFTHEEXOSKELETON-TYPE SYSTEM

The concept of the exoskeleton-type system is anextension of the exoskeleton in biology, in which theexoskeleton is a kind of external covering on an ani-mal to protect or support the creature, e.g. the shellof a crab. It serves not only as a protective coveringover the body, but also as a surface for muscle attach-ment, a water-tight barrier against desiccation, anda sensory interface with the environment. In a way,the metal full-body armour of knights can be called anexoskeleton as it provides a hard shell or skin to protectthe knight in a battle. Scientists, however, are extend-ing the idea even further. Exoskeletons now refer to

JMES936 IMechE 2008 Proc. IMechE Vol. 222 Part C: J. Mechanical Engineering Science

1600 C-JYang, J-F Zhang,Y Chen,Y-MDong, andY Zhang

Table 1 The comparisons between the exoskeleton technology and that concept in biology [5]

Function The biological exoskeleton The exoskeleton technology Application

Support Supporting the body of theinvertebrates

Supporting physically disabledpatient or walking assistance

Rehabilitation robotics or poweramplifier

Enhancement Enhancing the power of animals Strengthening the humanoperator

Assistance equipments

Protection Protecting the animals body Protecting the human operator Automatic armour for soldier,rescue devices or safe manip-ulation for the radioactivematerials in nuclear plant

Sensing and datafusion

Obtaining the information,acting as the sensorium

Interface of human operatorand the environment andmaking data fusion withinformation obtained by thehuman operator

Telerobotics, VR

supersuits or systems that expand or augment a per-sons physical abilities. They help a person lift or carryheavier loads, run faster, and jump higher. In the mili-tary, exoskeletons will help a soldier fight better sincehe can be better protected, carry more weapons andequipment, and have more strength than a normalperson has. The exoskeleton-type system makes fulluse of the human intelligence and the power of themachine to greatly enhance the performance of themanmachine system. Table 1 explains some analo-gies between the exoskeleton in the manmachinesystem and its concept in biology [5].

Basically, the control architecture of exoskeleton-type systems is quite different from the traditionalintelligent robotics. Figure 1 illustrates a typicalscheme of the exoskeleton-type system. In this con-trol architecture, the human operator is not only thecommander or the supervisor of the system, but alsoa part in the control loop, called man-in-the-loop. Inthe loop, the human operator mainly makes decisionsand the exoskeleton implements tasks. But feedbackinformation received by the human operator andexoskeleton keeps interchanging bilaterally betweeneach other. The human operator simultaneously actsas a feedback block in the looped control system. Thisis intuitive that the human operator receives the feed-back from environments and optimizes the control

Fig. 1 The scheme of the exoskeleton manmachineintelligent system

target. And if necessary, the exoskeleton can also takethe duty of data fusion, data processing, and evenassist the human operator with decisions. It puts moreemphasis on the combination of the human intel-ligence and machine power, and their informationexchanging, so that the man and the exoskeleton arecoupled together and both are irreplaceable. The intel-ligence of the machine system is enhanced while thepower of the human operator is also strengthened.Consequently, the exoskeleton-type system has higherintegrated intelligence and stronger adaptability tounstructured environments.

3 REVIEWOFTHEDEVELOPMENTHISTORY ANDCURRENT ADVANCEMENTS

In the mid-20th century, NASA details the backgroundof manmachine relationship between intimate part-ners in the report for large-scale space exploration.With the ebb of artificial intelligence (AI) in the1960s and 1970s, several researchers calmed down andthought out the research of AI. As the philosopherHubert Dreyfus argued in his book What comput-ers cant do? The limitation of artificial intelligence,traditional symbolic AI is a dead horse [6]. Dreyfuspresented four progressively weaker interpretations ofthe physical system hypotheses that he termed thebiological, psychological, epistemological, and onto-logical assumptions of traditional AI. At the end oflast century, many researchers realized that addinghuman intelligence is a potential alternative methodof addressing the problems faced by AI research upto a certain extent. Among them, Lu and Chen [1]from Zhejiang University put forward the concept ofhumachine [3], which indicates the specific methodto construct a manmachine system. The exoskeleton-type system is such a typical structure to realize thecouple of man and machine and set up their newrelationship in the human-centred intelligent system.

Proc. IMechE Vol. 222 Part C: J. Mechanical Engineering Science JMES936 IMechE 2008

Review of exoskeleton-type systems and their key technologies 1601

Basically, the development history of exoskeleton-type systems can be divided into three stages. Inthe early stage, exoskeleton-type systems were usu-ally used as the master-arm for telerobotics [79], thetools for human upper-limb or fingers posture mea-surement [1012], or the simple rehabilitation devicesfor physically disabled patients [13, 14]. With devel-opment of the force reflect and haptical feedbackresearch in the 1990s, the exoskeleton-type systemswith force feedback were widely employed in therobot bilateral teleoperation. As the force feedbackis added to the system, the human operator has thefeeling of doing the work directly instead of throughthe exoskeleton device, so that the performance ofthe telerobotic system is improved. This marks theexoskeleton technology coming into a new stage, inwhich the application of exoskeletons expands rapidlyto other fields, including power amplifier, hapticdevice in VR, rehabilitation, and so on. Most exoskele-tons completed or being studied in process sprangout. Nowadays, the exoskeleton technology is forg-ing ahead towards a new phase with special purposes,dedication, and high integration.

In the following sections, the exoskeletons by theirtwo prominent applications robot teleoperation andpower augmentation are introduced.

3.1 Exoskeletons in teleoperation

As mentioned above, an exoskeleton-type systemhighly integrates the sensing, data fusion, and datatransmission as well as force feedback, so that it cangenerate the realistic feeling as if the human operatoris doing the work directly. This transparent characterenhances the intuition of teleoperation and controlefficiency, especially in a different structure basedrobot masterslave control. The exoskeleton with forcefeedback is regarded as one of the best master arms intelerobotics.

In the 1970s, GE created the exoskeleton master sys-tem for robot teleoperation. Afterwards, the researchgroups from the University of Washington [15], OhioState University, and Stanford University also didsome related work [16, 17]. However these devicesare importable and need to be fixed on their bases asshown in Figs 2(a) and (b).

Compared with the importable exoskeleton-typemaster arm, another kind of the exoskeleton-typemaster arm is ungrounded with light weight. It isused by being seated on the shoulder of humanoperator. NASA and the Korea Institute of Scienceand Technology (KIST) have done some researchabout this kind of exoskeleton. In the KIST two setsof devices, pneumatic cylinders and electric brakes,were employed as force feedback actuators, as shown

Fig. 2 Exoskeleton-type master arm: (a) exoskeletonmaster of GE; (b) exoskeleton arm of Stanford[16]; (c) KIST exoskeleton arm with electricbrakes [18]

in Fig. 2(c). Furthermore, a 3RPS parallel mecha-nism was creatively introduced into the exoskeletonarm design for shoulder and wrist three-degree-of-freedom joints [1820]. Additionally, Bouzit et al. [21]integrated the Hall sensors for displacement measure,micro-actuators for force feedback, control circuitsonto a data glove, and developed the CyberGraspexoskeleton-type system (Fig. 3). This device is usedfor VR and can provide force feedback up to 16 N eachto the thumb, index, middle, and ring fingertips.

In our previous work [22, 23], a six-degree-of-freedom wearable exoskeleton arm, ZJUESA, forunderwater manipulators bilateral teleoperation wasdeveloped, as shown in Fig. 4. Because of the differ-ent structure between the master and slave manip-ulators, a general workspace mapping method wasproposed [24].

Specially, there is a kind of virtual exoskeleton inthis series. This work describes a vision-based humanmachine communication system that allows a com-puter or a control unit to see and track the motion

Fig. 3 The CyberGrasp exoskeleton-type system [21]

Fig. 4 The six-degree-of-freedom ZJUESA exoskeletonsystem



Fig. 5 The virtual exoskeleton [25]

of the upper limb of the human operator for somesimple teleoperation task, as shown in Fig. 5. Due toits specialty, a few researchers from the Institute forRobotics and Computer Science in Spain are work-ing in this field [25, 26]. They studied its applicationsin the teleoperation of underwater manipulators onremotely operated vehicles (ROV). However, the vir-tual exoskeleton is short of force and haptic feedback.

3.2 Exoskeletons for power augmentation

Body supporting and power amplification are twomain functions of the exoskeletons on animals.According to this principle, researchers all over theworld investigated a few types of exoskeletons aspower amplifier to help human operators finish thework with heavy loads. The robotics laboratory atUC Berkley designed a system called Human Pow-ered Extender to help workers on the assemblyline [27]. It decreases the burden on workers. Whenthe worker ports a load with 100 kg, with the helpof an exoskeleton-type system, he only feels a littleweight (about 5 kg) while the remaining 95 kg weightis supported by the exoskeleton-type system. Neuhauset al. [28] proposed concept designs for an underwa-ter swimming exoskeleton-type system. These designsare biologically inspired to strengthen the driver andimprove the poor maneuver. From references [29] and[30], a wearable power-assist device based on a curvedpneumatic muscle actuator (PMA) was designed. Itcan be directly put on like a coat because of itslightweight and soft texture.

In this field, the walking assistive exoskeleton (Fig. 6)is always the hotspot [31]. By combining the humanoperator and the robotic body together, it simplifiedthe challenges of gait stability of the automatic bipedwalking robot. The human operator plans the gait inreal-time, meanwhile the exoskeleton decreases theload of the human operator during walking and assistshuman to cover further distances for longer periodswith over-mounted loads on a physical interface basis.Table 2 lists the achievements of walking assistiveexoskeletons in the last century [32, 33].

Power-assisted exoskeletons have been developedsince 1948, when a Russian biomechanicist named

Fig. 6 Walking assistive exoskeleton: (a) HAL-5 system[34]; (b) BLEEX lower limb exoskeleton [31];(c) exoskeleton leg for humans walking poweraugmentation of Zhejiang University

Table 2 The history and achievements of walkingassistive exoskeletons in the last century[32, 33]

1948 The exoskeleton-type systemwith motor actuators designedby Professor Bernstein

1960 Hardyman system, designed byGE company. The soldier candrive it by its hydraulic systemto strengthen him and loadmore weapons

1970 Designed with 87 kg

1971Developed by Professor

Vukobratovic. It is used toassist patients for walking

1990 Gehhilfe system



Nicholai A. Bernstein thought up plans for an above-the-knee, electric-motor driven prosthesis to providemovement to casualties of war, but was never imple-mented. And then the first functional exoskeletonwas built by General Electric in 1968 and was calledthe Hardyman. It closely resembled the power loaderseen in the 1986 science fiction film Aliens, and washydraulically powered. The problem with the designwas that hydraulics required pumps and bladdersoccupied almost a room. The Russians experimentedwith a few more designs, but a Yugoslavian scientist,M. Vukobratovic, came up with the first anthropo-morphic exoskeleton to restore basic movement toparaplegics in 1972. A similar product named Gehhilfeby the German prosthetics company Otto Brock wasalso developed for the same reason in 1990.

In recent years, a major technological breakthroughhas taken place with the developments of relativedisciplines. The hybrid assistive limb (HAL) seriesdeveloped by Professor Yoshiyuki from Tsukuba Uni-versity was utilized to realize the waling aid for thegait disorder person [34]. Some sensors such as anglesensors, myoelectrical sensors, floor sensors, etc. wereadopted in order to obtain the condition of the HALand the operator. And it had hybrid control systems,which consist of the autonomous controller such asposture control and the comfortable power assist con-troller based on biological feedback and predictivefeed forward.

The Defense Advanced Research Projects Agency(DARPA) proposed a project, Exoskeleton for HumanPerformance Augumentation (EHPA). With the aimto strengthen the human operator, the system con-tained two hydraulic driven metal legs, an in-builtpower supply source and a backpack-like frame forload. Via the information detected by more than 40different sensors, the computer in the backpack-likeframe made the decision and steered the whole sys-tem. A hybrid power source supplied the energy forhydraulic actuators and control systems.

Additionally, the labs in NanyangTechnological Uni-versity and Kanagawa University have also reachedsome achievements in these years. The brief intro-ductions of these two systems can be found inreferences [35] and [36].

However, the key point of this kind of exoskeleton-type system is how to prospectively detect the motionof the human operator. This insight ensures theexoskeleton to follow the motion of a human operatorrather than counteract the human operator causingunnecessary resistance. Commonly, electromyogra-phy (EMG) signals on upper and lower limbs wereutilized to detect the motion tendency of the humanoperator. Some novel perspective control algorithmswere put forward, for example, the novel controlstrategy based on an adaptive network-based fuzzyinference system (ANFIS) controller was explored to

associate the planar pressure with the human operatorwalking gait on the exoskeleton leg [37].

Another exciting application of exoskeletons occursin the areas of physical disability and paralyzed reha-bilitation. Since MIT created the first rehabilitationexoskeleton arm for muscle dystrophy patients,Golden Arm and MIT-MAUS in 1970s and 1990s,respectively (Fig. 7(a)), Stanford University developedthe Arm Guide and MIME prototype systems in 2000and Rutgers University managed the researches on thelower limb exoskeleton rehabilitation system.

The Lokomat designed by Hocoma AG, Switzerland,was a bilateral robotic orthosis that was used in con-junction with a body-weight support system to controlpatient leg movements in the sagittal plane [3840].It can adjust the training gait style according to thebio-feedback of the patient, as shown in Fig. 7(b).Also HapticWalker was designed based on the prin-ciple of programmable footplates so that the patientsfeet were attached to two footplates, which moved hisfeet on arbitrary foot trajectories [41].

4 THE KEYTECHNOLOGIES OFEXOSKELETON-TYPE SYSTEMS

In this section, the key technologies in exoskeleton-type systems development are discussed. It includesbiomechanical design, system structure modelling,cooperation and function allocation between the manand exoskeleton, control strategy, and system safetyevaluation.

4.1 Biomechanical design

Exoskeleton-type systems are designed to invent amachine that could successfully track the humanmotions or assist the human to cover further distancesfor longer periods with over-mounted loads on a phys-ical interface basis. This translates into a number ofbiomechanical design specifications.

The exoskeleton should be anthropomorphic andergonomic, not only in shape but also in function. Itcan be very broadly described in terms of two classes:

Fig. 7 Rehabilitation exoskeleton-type systems: (a)MITMAUS exoskeleton arm; (b) Lokomat reha-bilitation exoskeleton [38]; (c) haptic walkingsystem [41]



Fig. 8 The six fundamental models of synovial joints[42]: (a) condyloid joint; (b) ball-and-socket joint;(c) pivot joint; (d) hinge joint; (e) planar joint; (f)saddle joint

1. The exoskeleton should be analogous to the humanlimbs and trunk in the case of joint positions anddistribution of degrees of freedom. As a result, itis important to investigate the atlas of the humanlimb and trunk during motions. Synovial joints arethe main components for human motion. In accor-dance with their forms of motions, they can becataloged into six fundamental joints [42] as theplanar joint, hinge joint, pivot joint, condyloid joint,saddle joint, and ball-and-socket joint [43] (Fig. 8).

Fig. 9 The motion model of human upper limb

Fig. 10 Design of exoskeleton leg structure from humanleg representative muscles for motion of kneeand hip

Figure 9 describes an example of the human upperlimb model based on the foregoing joint models.

2. The actuations in exoskeletons should be allocatedin the corresponding position to the representa-tive muscle in human limb and trunk, in order tosimulate the function of muscles during the pro-cess of the human operator moving [44]. As iswell known, the hip motion is generated by theextension/flexion of the rectus femoris and the glu-teus maximus, while the extension/flexion of thebiceps femoris and the vastus maximus producesthe extension/flexion of the knee. To adequatelysimulate the muscle activity during walking, for theaim of building an anthropomorphic leg, one actu-ator is adopted to simulate the function of the rectusfemoris and another to simulate biceps femoris, asshown in Fig. 10.

Additionally, the exoskeleton structure should belength adaptable. That is, all of the parts on theexoskeleton structure should be adjusted in a broadrange of length, thereby accommodating average peo-ple with different physiques. Under these circum-stances, the premise on which the exoskeleton isexpected to be ergonomic, highly maneuverable, andtechnically robust so the wearer can move, bend, andswing from side to side without noticeable reductionsin agility can be established.

4.2 System structure

To construct an exoskeleton-type system in a moremaneuverable and pragmatic way is still an open andchallenging issue. Comparing with traditional intelli-gent robot systems, exoskeleton-type systems have amuch closer connection with their human operators.In the working process, the exoskeleton-type system



makes full use of the intelligence of the human oper-ator and power of the machine, and combines thesemerits together. It also copes with the informationsensing and exchanging between the human operatorand the machine. Due to its specialization in struc-ture, it is hard to give out a universal exoskeleton-typesystem model.

In previous work, a model of humanmachine intel-ligent system was proposed [3]. In this model, thewhole system can be regarded as three main layers:sense, decision, and execution. The sense layer locatesat the top. It fuses the information data from humanoperator as well as machine sensors. The decisionlayer, according to the sensing information, organizesthe activities in the whole system, including humantasks and machine works. The bottom layer, namelythe execution layer, is responsible for the work imple-mentation, which can be finished by human handworkor auto-machine as the case may be. Definitely, theexoskeleton and its human operator have interactionwith each other in each layer. The relationship of thewhole system can be concluded as shown in Fig. 11.

Apparently the routine 1-2-3-4 represents that thework is operated only in hand while the routine 5-6-7-8 represents the work implementation by meansof automatic machine. The manmachine interfacebridges these two sites. In the exoskeleton-type sys-tems, the wearable exoskeleton acts as such an inter-face with its properties of friendliness, usability, trans-parency, etc. The machine is not only a tool butalso possibly an autonomous agent, and some coher-ence must be maintained between the humans andmachines actions on the environment. The humanand the machine become the complementary compo-nents of the system. Therefore, the capabilities of eachare exploited to their full capacity, thus achieving alevel of performance that neither could achieve alone.

4.3 Cooperation and function allocation

Function allocation between the human and machineraises another important issue for study and designof the exoskeleton-type system. The initial renownedcontribution to task allocation fields was made by Fittsin 1951 [45]. In his work, a solution was put forwardby attempting to decompose an activity on a generalbasis into elementary functions and to allocate eachone to the best efficient device, the human operatoror the machine, for that function. However, such anapproach was rapidly criticized because of its irra-tionality in the allocation process [46]. For example,according to the above-mentioned solution of Fitts,a task should be executed by the allocation of thedetection function to the machine, of the resolutionfunction to the human, and of the execution func-tion back to the machine. Obviously, this is not very

efficient. Also, as technology advances, the number offactors in which the abilities of humans exceed thoseof the machines decrease. If each task is allocated towhichever achieves the highest level of performanceat that function, the importance of the human in thesystem is reduced and their role becomes minimal.As a result of these criticisms, a number of develop-ments have occurred in the field of task allocation. Oneimportant change is viewing the human operator andmachine as complementary components of the sys-tem. This promotes both the human operator and themachine to achieve a level of performance that neithercould achieve alone.

However, there are some points that should benoted. The function allocation with complemen-tary concept is good for the exoskeleton-type sys-tem promotion. But it cannot ensure that humanoperators will not exert their responsibility over theentire exoskeleton-type system or even again takeup manual control. Their loss of expertise, becauseof the consequence of designing machines that playautonomous roles, either performing low-level func-tions or high-level functions, makes them likely toreach a poor performance. In addition, a few phe-nomena in exoskeleton-type systems have posed theawkward problem of shared supervision. More oftenthan not, the exoskeleton-type systems aim to reducethe workload of human operators when they are con-fronted with increasing requirements. The divisionof the supervision into two independent fields, thehuman operators and the exoskeletons field, gener-ates complacency and leads the intelligent machine,not alleviate the workload in terms of operators super-vision of the overall exoskeleton-type system [47].In response to complacency, the proposal of severalsolutions instead of only one from the machine is sug-gested. However, the decision of which suggestion tobe chosen or using manual mode is mainly affected bythe ratio between human operators trust in exoskele-tons and self-confidence. It is well believed that evenif the cooperative skills of the exoskeleton-type sys-tems are very restricted, exoskeleton-type systems canat least help the human operators to perform theactivities in better conditions than the current ones.

4.4 Man-in-the-loop control strategy

Some traditional manmachine intelligent systems areconcerned that the intelligent machines are designedto increase their adaptability to the variety of environ-ments, while they cannot equal operators in adaptiveskills [48, 49]. The main reason for these drawbacks isthe lack of real-time feedback to its human operatorwhen the machine is performing a task. This resultsin the well-known human-out-of-the-loop syndromeand leads the human operators, when necessary, to



Fig. 11 The diagram of manmachine intelligent system structure. Specification:The system struc-ture has three layers: sense, decision, and execution. In the sense layer, (1) the informationfrom environment to human; (19) the human sensing of the machine; (26) the humanself-sensing; (11) the human sensing of self-motion; (24) the feedback from the object to thehuman; (13) the data exchanging to manmachine interface; (5) the environment informa-tion measured by the machine; (20) the human state detected by the machine; (27) machineself-sensing; (12) the feedback from actuation; (25) feedback from the object; (14) data tothe manmachine interface. In decision layer, (2) the information from sensing organs;(15) the machine sensing information and machine processed data from manmachineinterface; (17) sending the thinking results to the manmachine interface; (6) sending thesensing information to the data processors; (16) the information of human sensing andthinking from the manmachine interface; (18) sending the data processing results to themanmachine interface. And in execution layer, (9) the human reflecting motion to theoutside motivation; (3) the motion commands from thinking organs; (21) human operat-ing the object through the machine; (4) human coping with the object in hand; (10) theactive signals to the actuations directly from the sensors; (7) the control signals from thedata processors; (8) machine working on the object; (22) the maintenance of the machine;(23) interaction from the machine

take control in hand without any clear situation aware-ness. Contrarily, in the exoskeleton-type system con-trol architecture as shown in Fig. 1, the exoskeletonreceives sequences of commands from the humanoperator or command generator of the system. In turn,the response of the environment is sensed and sig-nals are fed to the exoskeleton. By means of visual,audio, and tactile feedback, the human can intuitivelyand dynamically execute the work. It is the core thatis emphasized in the man-in-the-loop control archi-tecture [50]. However, it is not easy to fully or partlyrealize such an ideal control loop. It faces several dif-ficulties including the above-mentioned cooperationand function allocation between the human operatorand the exoskeleton, as well as the following dis-cussed information perception and fusion of thesetwo, real-time motion planning, safety control strat-egy with high-dependability, and so on. Dependabilityof the safety control strategy of the exoskeleton-typesystem calls for a modular and hierarchical architec-ture, which is also advantageous for testing the singlecomponents and isolating possible faults so as toachieve operating robustness. As known, exoskeleton-type systems are designed to assist the human oper-ator to work in dynamical environments. Dynamic

situations are uncertain since they are partially con-trolled. Unexpected factors can modify the dynamicsof the process and lead to the errors whose cost canbe very high in terms of accident and money. Hence,the human operators must manage the errors whenimplementing a decision. For them, the major risk isthe loss of control in a certain situation, acting toolate, although comprehension can reach a high levelby this time. Due to the need for continuous moni-toring, the response of the environment, exoskeletonoperation and the human operator state, as well as foronline changes in motion planning, the operation sys-tem of the control architecture on exoskeleton-typesystems must run in real-time. In many researcherswork, many methods for control in real-time could befound. The majority has involved tracking or flight-control tasks in which the operators were required todetect sudden failures in the dynamics of the con-trolled element [51]. Sheridan [52] proposed super-visory control of automated subsystems, primarilyin monitoring, diagnosis, and planning activities. Inthe work of Moncada et al., a multi-purpose manmachine interface for a control systems laboratory ispresented [53]. It possesses capabilities for multiplereal-time data acquisition, which is useful in practical



Fig. 12 The diagram of the event-based control archi-tecture

process identification and control. Suzuki et al. [54]developed a novel manmachine interface, which tookinto account the mutual cooperation between theoperator and an expert system in order to realize pre-cise decision-making and operation. But too manyfactors may affect the control performance, especiallythe delay in the LAN, which connects each module ofthe system for data exchange. The event-based con-trol architecture proposed by Kang et al. [55] and Tarnet al. [56] gives a new thought. It changes the continu-ous system into a discrete domain with event references instead of time reference t and it is proved to beeffective in dealing with the real-time control and goodsynchronization in the overall modules or hierarchiesin the control architecture [57].

Figure 12 shows the application of the event-basedcontrol architecture in the proposed exoskeleton-type master arm, ZJUESA, for the robot masterslavecontrol. The upper part in the dashed rectanglerepresents the slave robot arm and the other partexpresses the master system, namely exoskeleton-type system. These two sites over great distance areconnected through LAN. Each control command orfeedback information, for instance, the force feedback,is transmitted with a unique sequence s, which isa monotone increasing function of time t . Only theevent containing the latest sequence can pass throughthe event filter, which promises only one event existingat any time, and ensures that one control command iscorresponding to only one state feedback information.As a result, it leads the system to perform well in termsof good synchronization.

4.5 Information perception and data fusion

As mentioned above, in a manmachine system, theinformation interchange between the human opera-tor and the intelligent machine is necessary. In order to

achieve their bilateral natural communication, a uni-versal expression or even a kind of standard language isrequired. The formulized computer language is a pop-ular expression in the manmachine communication.Through this technology, the information can be easilyinput for controlling the behaviour of a machine, espe-cially a computer, but there is a barrier in the inverseapplication. Because of its inconvenience to be trans-lated into the natural human language, the feedbackinformation from the machine becomes incompre-hensive and may cause misinterpretation. Althoughthe visual, acoustic, graphic, and even tactile informa-tion have been added to enrich the language, yet it isstill far from satisfactory due to the limitation of rel-ative researches. The human language is also widelyproposed to be used in the manmachine communi-cation. However, a few difficulties should be faced [58].The arbitrariness of languages is the first problem. Thismeans that there is no logical connection betweenmeanings and sounds. This is a sign of sophistica-tion and it makes it possible for languages to havean unlimited source of expressions. Second, languagesare productive or creative in that it makes possible theconstruction and interpretation of new signals by itsusers. This is why an infinitely large number of sen-tences or messages can be produced, including muchof what they have never heard before. Additionally,languages are rich. Sometimes for the same expres-sion, a wide choice of words can be offered in onelanguage. These natures of human languages block itsapplication in the manmachine communication.

At present, some novel bio-chips and bio-sensorsbased on the EMG, which evaluates and records phys-iologic properties of muscles, electroencephalograph(EEG, which monitors brain waves), and electro-oculograph (EOG, which monitors eye movement) canbe found in other references to assist the exoskeleton-type systems to detect the motion of the humanoperator [34, 5962]. Nevertheless, most of thesedevices still stay in the labs for experiments rather thanpractice, in which top signal processing technologiesand advanced prospecting algorithms are required.Certainly, these bio-techniques will be widely appliedon exoskeletons in near future. For example, in neu-rotechnology, a biochip (DNA microarrays, proteinchips, RNA chips) may be embedded in the humanbrain, which allows him/her to control a computer ora machine using his thoughts.

In fact, only the implementation of the informa-tion perception and interchange between the humanoperator and the machine in such a manmachineintelligent system is not enough. Depending on themulti-sensor system and network, the perceptioninformation should be processed and synthesizedbased on the knowledge in the expert database.As this work involves the data fusion between thehuman operator and the machine, some challenges



Fig. 13 The exoskeleton fuzzy neural-network control architecture for manmachine informationperception

are encountered, for example, the information expres-sion (how to express the information so that it canbe comprehended by both human and machine),symbol meanings (how to define the symbols in theinformation), and information management (how tomanage the information exchange and modify theexperts database) [63]. Since humans evaluate inputfrom their surroundings in a fuzzy manner, the fuzzylogic can be recognized as an ideal medium. In pre-vious important work, the Mamdani fuzzy logic wasintroduced in the man-in-the-loop based informationfusion for exoskeleton-type systems [64, 65]. Equation(1) expresses this concept in fuzzy implication

if x is Ai and y is Bi, then z is Ci (1)

It also can be represented in mathematics as

Ai(x) Bi(y) Ci(z) (2)

where Ai is the fuzzy information from the human, Bithe accurate information from the exoskeleton, andCi the results of the information fusion and synthe-sis are the fuzzy sets associated with the universeof discourse X , Y , and Z , with membership func-tion denoted Ai (x), Bi ( y), and Ci (z), which can beexpressed as

Ai = {(x, Ai (x)) : x X } Bi = {(y, Bi (y)) : y Y }Ci = {(z, Ci (z)) : z Z} i = 1, 2, . . . , n (3)

Note that after the fuzzification, the informationfusion problem between the human and the machine

becomes a routine fuzzy logic case. With an appropri-ate rule base, it is easy to get the results by defuzzifi-cation. Here, the neural-network control architectureis introduced to specify the rule base, as depicted inFig. 13. It provides a formal methodology for represent-ing and implementing a humans heuristic knowledge.It allows for the modification of parameters and learn-ing fuzzy rules from input/output data pairs, incorpo-rating prior knowledge of fuzzy rules, fine tuning themembership functions, and acting as a self-learningcontroller by automatically generating the fuzzy rulesneeded.

4.6 System safety and reliability

As mentioned, the exoskeleton-type systems alwayswork around humans or even touch with them andthe conventional safety strategies for industrial robotsare not well fit to exoskeleton-type systems. Consid-ering the impact force in a sudden collision, which isthe possibility of injury to humans from exoskeleton-type systems, Ikuta et al. [66] proposed a generalsafety evaluation method by employing the criticalimpact force Fc as a minimal impact force that causesinjury to humans and giving the definition of danger-index as the producible impact force F against Fc inequation (4)

= FFc

0 (4)

By using this method, not only the impact force Fbut also many types of safety or dangerousness canbe evaluated quantitatively. Definitely, minimizing the



impact force by means of safe design, control strategyis the effective way to reduce the damage of the humanfrom the collision.

4.6.1 Reducing the weight and inertia of exoskeletonpart

According to Newtons law of motion, the relationsbetween the danger index and the weight and inertiarespectively, as shown in equation (5), can be obtained

={

ma/Fc linear motion

I /lFc rotary motion(5)

where m and I are the mass and inertia of the exoskele-ton part, respectively; a and are system accelera-tion and angular acceleration, respectively; and l thedistance from joint to contact point.

4.6.2 Distributed macro-mini actuation and jointcompliance (DM2)

Another approach to reduce the manipulators arminertia for safety, while preserving performance, isthe methodology of distributed macro-mini actuation(DM2) [67]. Conventionally, in an exoskeleton-type,a pair of actuators is employed for each degree offreedom. The DM2 actuation approach is to dividethe torque generation into separate low- and high-frequency actuators. The low-frequency actuations,where large actuators are required, are located at thebase of the exoskeleton, while the high-frequencytorque actuators, often small motors, are placed at thejoints to ensure the control performance. This reducesthe overall inertia to a certain degree while decreasingthe overall weight of the motion part in the system.

Additionally, an effective strategy for enhancing thesystem safety by reducing the impact force in colli-sion is to have compliance in each joint as shown inFig. 14(b). With the rotary joint on exoskeletons, theimpact force F in collision can be calculated as

F = I ( )

l dt(6)

And according to the collision dynamical equation ofthe system

I + C + K = 0 (7)

The damage index of the system can be derived as

= I Fc dt

(8)

where dt = (tan1(n/d) + /2)/d , and n =K /l,d = n

1 2, = C/2IK .

Fig. 14 Safety design of exoskeleton [68]: (a) with smallmass and low inertia; (b) with compliant joints

4.6.3 Mounting stop block

For ensuring safety, the stop block is important tothe system. The joint motion spaces on exoskele-tons should be limited in those of human operator.Especially under the abnormal state, the exoskeletoncannot hurt the operator due to its over-scaled jointmotion spaces. Generally the stop blocks on intelli-gent machines can be cataloged into soft and hardstop blocks, which are implemented in software andhardware, respectively. According to the definition ofdamager index, it can be concluded as equation (9)

=

0 with stop block

f /Fc without stop block and f < Fc1 without stop block and f Fc

(9)

where f is the impact force to the human operator dueto the over motion of the exoskeleton joint.

4.6.4 Fault hanging

To preserve the safety of humans interacting withexoskeletons, especially in occurrence of unexpectedevents, as failures or abrupt changes of the environ-ment, fault handling and fault tolerant control haveto be considered as fundamental functionalities. Itrequires the application of a sequence of activities fordealing with faults [68, 69]:

(a) fault prevention, to prevent the occurrence orintroduction of faults;

(b) fault removal, to reduce the number and severityof faults;

(c) fault detection and isolation, to recognize theoccurrence of a fault and characterize its type;

(d) fault tolerance, to avoid service failure in presenceof faults;

(e) fault forecasting, to estimate the present number,the future incidence and the likely consequencesof faults.

Fault prevention and fault removal are collectivelyreferred to as fault avoidance. Fault detection andisolation are parts of fault diagnosis based on theexisting analytical fault diagnosis techniques includ-ing observe-based approaches, parameter estimation



techniques, and so on. Fault tolerance and fault fore-casting are collectively referred to as fault acceptancethat may guarantee that the effects of local faultsremain internal to the modules, and also permits thereconfiguration of the system.

5 CONCLUSIONS

The exoskeleton-type system is designed as thehuman centered manmachine system. It fully com-bines the merits of the human control system androbotic power. The potential exists for the exoskeleton-type system to establish a technology revolution thatmay exceed the impact of intelligent mechatronics androbotics on society. Its two prominent applications areteleoperation and power amplification. As the field ofexoskeleton-type systems is adopted by many disci-plines, various advantageous mechanical and controlproperties are exploited, the diversity and acceptanceof exoskeleton-type system will grow. However, it isdifficult to develop a quite excellent exoskeleton-typesystem due to several technological limitations andchallenges.

Biomechanical design is a main research on thedesign method of exoskeleton based on the anatomy ofhuman limb movement. Although, many structures ofexoskeletons have been proposed, it has proved to bedifficult to mechanically emulate the natural motionof the human. Thus, the structures of exoskeletonsshould be biomechanically investigated not only inshape but also in function, and in degrees of freedomdistribution as well.

System structure modelling is severely concernedwith the cooperation and function allocation betweenhuman operator and exoskeleton in the sense, deci-sion, and execution layers. There must be some rulesfor building good cooperation between them, and allo-cating functions to the one or the other that is the mostappropriate for the job. Definitely, it is well believedthat even if the cooperative skills are very restrictedby recent technology, exoskeletons can at least helpthe human operator to perform activities in betterconditions than the current ones.

Control strategy also raises an important issue. Dif-ferent from the autonomous robotics, the exoskeleton-type systems focus on the man-in-the-loop controlstrategy. But it is not easy to fully or partly realizesuch an ideal control loop. Cooperation and func-tion allocation, manmachine information exchange,real-time motion planning and safety control are thedifficulties faced by building such a control strategy.

Safety evaluation of exoskeleton-type systemsshould also be considered carefully. Exoskeletons arecloser to human and its security should be completelypromised. However this point is rarely discussed in

existing research. Few relative literature, design stan-dards or specific law files could be found, that maylead to a fatal mistake.

Indeed, significant on-going efforts in advancedbiomechanical and automatic engineering areimproving the performances of exoskeleton-type sys-tems. In the near future, exoskeleton-type systemsmay play a tremendously significant role in the never-ending development of the manmachine system.

ACKNOWLEDGEMENTS

This project is funded by the National Natural ScienceFoundation of China (No. 50305035).

REFERENCES

1 Lu, Y. X. and Chen, Y. Foundation of the HumachineSystem (in Chinese). Chin. J. Mech. Eng., 1994, 30(6), 19.

2 Yang,C. J.,Chen,Y.,and Lu,Y. X. Study on the humachineintelligent system and its application (in Chinese). Chin.J. Mech. Eng., 2000, 36(6), 4247.

3 Yang, C. J. Study on the theory of humachine intelli-gent system and its application. PhD Thesis, ZhejiangUniversity, Hangzhou, 1997.

4 Available from http://www.wired.com/wired/images.html? issue=13.01&topic=ironmen&img=1

5 Li, X. M. Teleoperation of internet based robot using ahuman arm exoskeleton. PhD Thesis, Zhejiang Univer-sity, Hangzhou, 2004.

6 Dreyfus,H. L.What computers cant do? the limitation ofartificial intelligence, 1979 (Harper & Row, HarperCollins,New York).

7 Repperger, D. W., Remis, S. J., and Merrill, G. Perfor-mance measures of teleoperation using an exoskeletondevice. In Proceedings of the IEEE International Confer-ence on Robotics and Automation, Nice, France, 1990,vol. 27, pp. 552557.

8 Repperger,D.W. and Remis,S. J.Use of a multi-axis Fittslaw paradigm to characterize total body motion-a studyin teleoperation. In Proceedings of the IEEE Confer-ence on System Engineering, Pittsburgh, PA, USA, 1990,pp. 105108.

9 Jau,B.M.Anthropomorphic exoskeleton dual arm/handtelerobot controller. In Proceedings of the IEEE Work-shop on Intelligent Robots and systems, Tokyo, Japan,1988, pp. 715718.

10 Sala, R., Milanesi, S., and Rovett, A. Measurement ofsingle phalanges positioin: a new fast and accurate solu-tion. In Proceedings of the 20th International Conferenceon Industrial Electronics, Control and Instrumentation,Bologna, Italy, 1994, vol. 2, pp. 942945.

11 Wright, A. K. and Stanisic, M. M. Kinematic map-ping between the EXOS handmaster exoskeletonand the Utah/MiT dextrous hand. In Proceed-ings of the IEEE International Conference onSystem Engineering, Pittsburgh, PA, USA, 1990,pp. 101104.



12 Giuffrida, F., Piaggio, M., and Guerrasio, C. G-EXO, amodular exoskeleton as multi purpose multi media inter-face. In Proceedings of the IEEE International Workshopon Robot and Human Communication, Tsukuba, Japan,1996, pp. 213216.

13 Brown, P., Jones, D., Singh, S. K., and Rosen, J. M.The exoskeleton glove for control of paralyzed hands.In Proceedings of the IEEE International Conference onRobotics and Automation, Atlanta, GA, USA, 1993, vol. 1,pp. 642647.

14 Lee, S., Agah, A., and Bekey,G. IROS: an intelligent reha-bilitative orthotic system for cerebrovascular accident.In Proceedings of the IEEE International Conference onSystems, Man and Cybernetics, Los Angeles, CA, USA,1990, pp. 815819.

15 Rosen, J.,Hannaford,B., and Burns, S. Neural control ofan upper limb powered exoskeleton type system-grantreport. In the First NSF Robotics and Computer Vision(RCV)Workshop, LasVegas, Nevada, 2627 October 2003,pp. 2123.

16 Turner, M. L., Findley, R. P., Griffin, W. B., Cutkosky,M.R., and Gomez, D. H. Development and testing of atelemanipulation system with arm and hand motion. InProceedings of the ASME Dynamic Systems and ControlDivision, Orlando, FL, 2000, pp. 10571063.

17 Koyamal, T., Yamano, I., Takemura, K., and Maeno, T.Multi-fingered exoskeleton haptic device using passiveforce feedback for dexterous teleoperation. In Proceed-ings of the IEEE/RSJ International Conference on Intelli-gent Robots and Systems, Lausannne, Switzerland, 2002,pp. 29052910.

18 Kim, Y. S., Lee, J., Lee, S., and Kim, M. A force reflectedexoskeleton-type masterarm for humanrobot interac-tion. IEEE Trans. Syst. Man Cybern. A, Syst. Humans, 2005,35(2), 198212.

19 Park, J. O.Virtual reality and teleoperation in robotics. InProceedings of the 3rd IWAN, 1999, pp. 7890.

20 Jeong, Y. K., Lee, D. J., Kim, K. H., and Park, J. O.A wearable robotic arm with high force-reflection capa-bility. In Proceedings of the IEEE International Workshopon Robot and Human Interactive Communication, 2000,pp. 411416.

21 Bouzit, M., Burdea, G., Popescu, G., and Boian, R.The Rutgers Master II-new design force-feedback glove.IEEE/ASME Trans. Mechatronics, 2002, 7(2), 256263.

22 Zhang,J. F.,Yang,C. J.,andChen,Y.Exoskeleton arm withforce feedback for robot bilateral teleoperation. Prog.Nat. Sci., 2007, 17(9), 5662.

23 Zhang, T. and Nakamura, M. Neural network-basedhybrid human-in-the-loop control for meal assistanceorthosis. IEEE Trans. Neural Syst. Rehabil. Eng., 2006,14(1), 6475.

24 Chen, Y., Zhang, J. F., Yang, C. J., and Niu, B. Theworkspace mapping with deficient DOF space for thePUMA 560 robot and its exoskeleton-arm by usingorthogonal experiment design method. Robot. Comput.-Integr. Manuf., 2007, 23(4), 478487.

25 Amat, J., Frigola, M., and Casals, A. Virtual exoskeletonfor telemanipulation. Lecture notes in control and infor-mation sciences. In Experimental RoboticsVII (Eds D. Rusand S. Singh), 2001, pp. 2131 (Springer-Verlag, London,UK).

26 Amat, J., Casals, A., Frigola, M., and Martin, E. Exper-imental. Bilateral control telemanipulation using a vir-tual exoskeleton. Springer Trans. Adv. Robot., 2003, 5,350361.

27 Available from http://www.me.berkeley.edu/hel/hydextender.htm

28 Neuhaus, P. D., Sullivan, M. O., Eaton, D., Carff, J., andPratt, J. E. Concept designs for underwater swimmingexoskeletons. In Proceedings of the IEEE InternationalConference on Robotics and Automation, 2004, pp. 48934899.

29 Gao, L., Noritsugu, T., Takaiwa, M., et al. Develop-ment of wearable power assist device using curvedpneumatic artificial rubber muscle. In Proceedingsof the International Conference on Fluid PowerTransmission and Control, Hangzhou, China, 2005,pp. 330334.

30 Naruse, K., Kawai, S., Yokoi, H., and Kakazu, Y. Devel-opment of wearable exoskeleton power assist sys-tem for lower back support. In Proceedings of theIEEE/RSJ International Conference on Intelligent Robotsand Systems, Las Vegas, Nevada, USA, 2003, vol. 3,pp. 36303635.

31 Zoss, A. B., Kazerooni, H., and Chu, A. Biomechani-cal design of the Berkeley lower extremity exoskeleton(BLEEX). IEEE/ASME Trans. Mechatronics, 2006, 11(2),128138.

32 Niu, B. Study on the design and control of a wearableexoskeleton leg for humans walking power augmenta-tion. MSc Thesis, Zhejiang University, Hangzhou, 2006.

33 Vukobratovic, M., Borovac, B., Surla, D., and Stokic, D.Biped locomotion: dynamics, stability, control, and appli-cation, 1990 (Springer-Verlag, Berlin).

34 Available from http://sanlab.kz.tsukuba.ac.jp/HAL/indexE.html.

35 Liu, X. P. and Low, K. H. Development and prelim-inary study of the NTU lower extremity exoskeleton.In Proceedings of the IEEE Conference on Cybernet-ics and Intelligent Systems, Singapore, 2004, vol. 3,pp. 12431247.

36 Yoshimitsu, T. and Yamamoto, K. Development of apower assist suit for nursing work. In the SICE 2004Annual Conference, Hokkaido, Japan, 2004, pp. 577580.

37 Yang, C. J., Niu, B., and Chen, Y. Adaptive neuro-fuzzycontrol based development of a wearable exoskeletonleg for human walking power augmentation. In Pro-ceedings of the IEEE/ASME International Conference onAdvanced Intelligent Mechatronics, United States, 2005,pp. 467472.

38 Colombo, G., Joerg, G., Schreier, R., and Dietz,V. Tread-mill training of paraplegic patients using a roboticorthosis. J. Rehabil. Res. Dev., 2000, 37, 693700.

39 Riener, R., Luenenburger, L., Jezernik, S.,Anderschitz, M., Columbo, G., and Dietz, V. Patient-coorperative strategies for robot-aided treadmilltraining: first experimental results. IEEE Trans. NeuralSyst. Rehabil. Eng., 2005, 13(3), 380394.

40 Riener, R. Control of robots for rehabilitation, 2005,pp. 3336 (EUROCON, Belgrade, Serbia & Montenegro).

41 Schmidt, H., Piorko, F., Bernhardt, R., Kruger, J., andHesse, S. Synthesis of perturbations for gait rehabilita-tion robots. In Proceedings of the IEEE 9th International



Conference on Rehabilitation Robotics, Chicago, IL, USA,2005, pp. 7477.

42 Tortora,G. J. and Grabowski, S. R. Principles of anatomyand physiology, 9th edition, 2000 (John Wiley, New York).

43 Premkumar, K. The massage connection: anatomy andphysiology, 2004, (Lippincott Williams & Wilkins, Mary-land).

44 Zoss, A. B., Kazerooni, H., and Chu, A. Biomechani-cal design of the Berkeley lower extremity exoskeleton(BLEEX). IEEE/ASME Trans. Mechantronics, 2006, 11(2),128138.

45 Fitts,P.M.Human engineering for an effective air naviga-tion and traffic control system, 1951 (National ResearchCouncil, Washington, DC).

46 Older, M. T., Waterson, P. E., and Clegg, C. W. A crit-ical assessment of task allocation methods and theirapplicability. Ergonomics, 1997, 40(2), 151171.

47 Hoc, J. K. From human-machine interface to human-machine cooperation. Ergonomics, 2000, 43(7), 833843.

48 Mosier, K. L. Myths of expert decision-making andautomated decision aids. In Naturalistic decisionmaking, pp. 319330 (Lawrence Erlbaum Associates,Mahwah, NJ).

49 Parasuraman,R. Humans and automation: use, misuse,disuse, abuse. Hum. Factors, 1997, 39, 230253.

50 Zhang, J. F., Yang, C. J., and Chen, Y. Use orthog-onal experimental method to the optimal design forexoskeleton arm. WSEAS Trans. Syst., 2007, 6(6), 10951101.

51 Kessel, C. and Wickens, C. D. The transfer of fail-ure detection skills between monitoring and controllingdynamic systems. Hum. Factors, 1982, 24, 4960.

52 Sheridan, T. B. On how often the supervisor shouldsample. IEEE Trans. Syst. Sci. Cybernet., 1970, SSC-6,140145.

53 Moncada, R. S. Ferrer, J. S., Pavone, C., and Vera, M.A multi-purpose real-time man-machine interface forcontrol systems laboratory. Comput. Ind. Eng., 1997,33(1), 425428.

54 Suzuki, T., Sakino, Y., Nakajima, M., Asama, H., Fujii,T., Sato, K., Kaetsu, H., and Endo, I. A novel man-machine interface for a bioprocess expert system con-structed for cooperative decision making and operation.J. Biotechnol., 52(3), 277282.

55 Kang, W., Xi, N. and Tan, J. D. Analysis and design ofnon-time based motion controller for mobile robots. InProceedings of the 1999 IEEE International Conferenceon Robotics and Automation, Detroit, Michigan, USA,1999, pp. 29642969.

56 Tarn, T. J., Song, M., and Xi, N. Intelligent planning andcontrol for hybrid system. In Proceedings of the 1998

IEEE/RSJ International Conference on Intelligent Robotsand Systems, Victoria, BC, Canada, 1998, pp. 972977.

57 Li, X. M.,Yang, C. J., Chen,Y., and Hu,X. D. Hybrid eventbased control architecture for tele-robotic systems con-trolled through Internet. J. Zhejiang Univ. (Sci.), 2004,5(3), 296302.

58 Wyard, P. Representational issues for context-free gram-mar induction using genetic algorithms, 1994 (Springer-verlag, Berlin).

59 Vogel, G. Part man, part computer: research tests thelimits. Science, 2002, 295(5), 1020.

60 Fleischer, C., Wege, A., Kondak, K., and Hommel, G.Application of EMG signals for controlling exoskeletonrobots. Biomed. Tech., 2006, 51, 314319.

61 Ji, J., Yu, M. S., Zhou, Y. B., and Jin, Z. R. Awireless EEG sensors system for computer assisteddetection of alpha wave in sleep. In Proceedings ofthe 2005 IEEE Engineering in Medicine and Biol-ogy 27th Annual Conference, Shanghai, China, 2005,pp. 53515353.

62 Kiguchi, K., Esaki, R., Tsuruta, T., Watanabe, K.,and Fukuda, T. An exoskeleton for human elbowand forearm motion assist. In Proceedings of theIEEE/RSJ International Conference on Intelligent RobotsSystems, Las Vegas, Nevada, USA, 2003, vol. 3,pp. 36003605.

63 Lambert, D. A. Grand challenges of information fusion.In Proceedings of the 6th International Conference ofInformation Fusion, Cairns, Queensland, Australia, 2003,vol. 1, pp. 213220.

64 Yang, C. J. and Chen, Y. Study the model of the com-prehensive perception of humachined intelligent sys-tem (in Chinese). Control Theory Appl., 2000, 17(2),220224.

65 Passino, K. M. and Yurkovich, S. Fuzzy control, 1998(Addison-Wesley Longman, California).

66 Ikuta, K. and Nokata, M. Safety evaluation method ofdesign and control for human-care robots. Int. J. Robot.Res., 2003, 22, 281297.

67 Zinn,M., Khatib, O., Roth, B., and Salisbury, J. K. A newactuation approach for human-friendly robot design. InProceedings of the International Symposium on Exper-imental Robotics, S. Angelo dIschia, Italy, July 2002,pp. 379398.

68 Avizienis, A., Laprie, J. C., Randell, B., and Landwehr, C.Basic concepts and taxonomy of dependable and securecomputing. IEEE Trans. Dependable Secur. Comput.,2004, 1(1), 1133.

69 De Santis, A., Siciliano,B.,De Luca,A., and Bicchi, A. Anatlas of physical humanrobot interaction. Mech. Mach.Theory, 2008, 43(3), 253270.


Documents

A Review of Exoskeleton-type Systems and Their Key Technologies 2008