TMECH 06 2023 15840 - Proof - Hi

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

IEEE Transactions on Mechatronics

Integrated Observer-based Prescribed Performance Control


for Back-Support Exoskeleton with Time Delays

Journal: Transactions on Mechatronics

Manuscript ID TMECH-06-2023-15840

Manuscript Type: Regular paper

Date Submitted by the


30-Jun-2023
Author:

Complete List of Authors: Li, XiaoGang; Tianfu Xinglong Lake Laboratory,


Ke, Yuan; Tianfu Xinglong Lake Laboratory
He, Jiang; Tianfu Xinglong Lake Laboratory

Automotive control < Automotive systems, Applications (robotics) <


Robotics, Biomimetic & bio-inspired robotics < Robotics, Unmanned
Keywords:
autonomous systems < Robotics, Linear & nonlinear control <
Intelligent control

Are any of authors IEEE


No
Member?:

Are any of authors ASME


No
Member?:
Page 1 of 11 IEEE Transactions on Mechatronics

1
2
3
4
5
Integrated Observer-based Prescribed Performance
6
7 Control for Back-Support Exoskeleton with Time
8
9
10
Delays
11 XiaoGang Li1, Yuan Ke1, Jiang He1
12
1. Tianfu Xinglong Lake Laboratory
13
14 Abstract—Lower back pain and lumbar muscles strain are that WBE (both rigid and soft) can reduce back extensor
common seen in different age. For this kind of chronic disease, activity, muscle fatigue, and increase worker comfort and
15
back-support exoskeleton(BSE) provides a non-invasive productivity [2]-[4]. As a new type of paramedical device, the
16
therapeutic schemes. The BSE supports the upper body BSE can effectively help patients suffering from back pain to
17
through the thrust of its multiple support bars, thus reducing relieve their symptoms as well as assist in post-operative
18 the stress on the lumbar spine and lumbar muscles and thus recovery through its support role. Active BSEs can be applied
19 providing relief for lumbar spine disorders. As a brace for to different movements and work situations as they can reduce
20 upper body, the BSE have to closely touch the body, so the the level of activation of the back muscles through their
21 contacted part is always composed of materials with low bracing effect, thus achieving a reduction in the risk associated
22 stiffness or with some buffer unit to protect the body, which with lumbar spine disorders, the aetiology of which is
23 brings time delay effect on the supporting force. BSE is a characterised by repetitive lifting and/or uncoordinated and
24 many-in-mangy-output system, the disturbance, uncertainties static postures [5]-[7].
25 and time delay from different dimension would produce Although both are designed to assist, support and relieve
26 interaction effect, which brings challenge in the control system. force, BSEs can be divided into two different types according
27 Therefore, this paper is devoted to designing a new prescribed to their principle of operation: passive and active exoskeletons
28 performance controller to deal with the time delay effects and [8]-[9]. The passive exoskeleton relies on the restoring force
29 match the control goal, namely, lower overshoot, fast response of springs or elastic materials to provide tension or support to
30 process and high precise. Additionally, to address the modeling the patient and has the advantage of being lightweight, simple
31 errors and external disturbance, a new integrated observer and economical, but it is difficult to adjust the compensation
32 composed of linear extended state observer and sliding mode force required by the wearer in real time [10]-[12]. The study
33 observer is designed to estimate the lumped perturbation then suggests that passive exoskeletons may be preferable for static
34 compensate it. tasks, while active exoskeletons may be more advantageous for
35 Key words: back-support exoskeleton, sliding mode dynamic tasks [9], [13]. Therefore, in order to improve the
36 observer, linear extended state observer, parallel manoeuvrability as well as the adaptability of exoskeletons,
37 mechanism, prescribed performance controller. researchers and engineers tend to develop more effective and
I. Introduction versatile active solutions. For example, in [14],
38
People who have to perform single repetitive movements, magnetorheological fluid is used to design a semiactive device,
39
such as long-time sitting, lifting and carrying, etc., are prone to it work well but asks the participate to stand stably due the
40
lumbar spine disease. Both those who work by setting in office support force is main from the thigh. Reference [15] designs a
41 novel controller that uses a single adaptive dynamic motion
42 and those who work outdoors with physical effort are at risk of
low back problems. According to Liberty Mutual's 2016 primitive to facilitate a variety of lifting motions for wearable
43 robots to assist with discrete movements. Using a linear drive
44 statistics, back injuries accounted for 17.3% of all injuries in
the USA and the highest proportion of work-related mechanism based on a single tandem elastic actuator, literature
45 [16] proposes an exoskeleton consisting of a ball screw driver
46 musculoskeletal disorders [1]. These diseases are usually
treated by surgery, or mechanical traction, while the surgery is and a tandem elastic actuator, allowing for better power flexion
47 control manoeuvrability while maintaining the natural walking
always with risk and traction equipment is too expensive for
48 advantage of almost zero impedance. As can be seen from the
some family to afford it. Therefore, a kind of lightweight,
49 available review literature, there have been many
economical, non-invasive assistive device is what many
50 patients need today. Driven by this objective need, the back- developments in control schemes for exoskeletons, thanks to
51 support exoskeleton (BSE) was born to provide a new solution advances in control algorithms [17]-[20]. Due to the
52 for a new type of complementary therapy. exoskeletons need to touch body directly, the support point and
53 BSE are expected to improve ergonomics by reducing the brace ring cannot be built by completely rigid materials, which
54 load on the lumbar spine. Back-support exoskeletons are brings time delay phenomenon to the drive force, meanwhile,
55 designed to reduce strain, fatigue and injury risk for bending, the inertial measurement unit also cannot track the real time
56 extending and lifting workers. There is substantial evidence posture perfectly because its limited data-update-rate [21]. The
57
58
59
60
IEEE Transactions on Mechatronics Page 2 of 11

1 time delay phenomenon in the exoskeleton is therefore of great data and send it to electric fastening kit to achieve automatic
2 importance in order to improve control accuracy and thus adjustment of tension on belt.
3 enhance wearer’s comfort.
4
Pressure sensor for
Upper electric
Control systems for exoskeleton are still in the fastening kit
Lower chest
supporting platform
lower chest

5 development stage and there are a few control solutions for


6 them with time lag, which can be divided into three categories: Control card
Battery

7 1) Compensate for time delay processes by using mathematical Switch Electric push rod

8 models that are as accurate as possible to invert them [22]. 2)


9 Approximation of the time lag process by means of an Manual
10 estimator or reinforcement learning to resist compensation of fastening kit Lumbar curve
adaptation device
11 the time lag effect[23], [24]. 3) Time delay are treated directly Spring buffer

12 as perturbations and then feed-forward compensation is Lower electric Pelvic supporting Pressure sensor for

13 performed after designing observers to estimate them [21],


fastening kit platform pelvic

Figure 1 Physic structure of the back-support exoskeleton


14 [25], [26]. However, it understandable that the exact a3
15 mathematical model is different to obtain and the neuro a4 z
y
16 network needs large amount computation which bring m
R04
o
bw
17 challenge to micro CPU, as well as needs long-time and R03 x
a2
bl
18 uncomfortable wear to obtain the training data to modify the a1
19 network of those algorithms. Additionally, it is feasible to treat
20 the slight time delay as internal perturbation and then design l4
l3
disturbance observer to estimate and compensate it, while this
21
scheme would not work in the appearance of the large scale l0
22 δ4
delay[27]. l1 δ3
23 δ0 l2
BSE, a MIMO system with strong coupling effects, Z R32
24 constrained by its complex mechanical structure thus is
A4 R42 Y A3 R31
Fp

25 different to obtain the exact dynamic and kinematic model.


aw R01
O c
26 Meanwhile, the whole structure should be light which ask the
δ1 B R02 X
δ2
27 computation system cannot be overweight, thus the hardware
R12
A1 R11 al R21
28 is different to afford the high-performance computation for
A2 R22
Figure 2 Geometric construction diagram of the exoskeleton
29 neuro network. Forward and model-free prescribed The exoskeleton is powered by the thrust from the top to
30 performance controller (PPC) would be an efficient way to the bottom push rod, but given that the dynamics analysed in
31 address this control problem [33], while the research about the motion control is the relative motion of the top and bottom
32 PPC in MIMO system is relatively less, therefore, this paper platforms, we can switch the top and bottom platforms to
33 tries to design a new PPC and apply it into BSE to address the facilitate the kinematic analysis and design of the controller, as
34 time delay effect and enhance the tracking accuracy. shown in Figure 2. Four side chains connect the base platform
35 Additionally, to deal with the modeling errors and external B with universal hinges and support the moving platform m
36 disturbance, an integrated observer composed of linear with spherical hinges, where the intermediate piston cylinder
37 extended state observer and sliding mode observer is designed (active rods) and follower rods act as prismatic pairs to provide
38 to estimate the lumped perturbation then compensate it. power. The centres of the four spherical hinges ai (i=1, 2, 3, 4)
39 The remaining parts of this paper is arranged as follows: on the moving platform form a rectangle with sides bl, width
40 Section II is devoted to analyzing the physical structure and bw and centre o. The centres of the four ball hinges Ai (i=1, 2,
41 discussing the detailed kinematic model and dynamic model. 3, 4) on the base platform also form a rectangle with length al,
42 The building process of controller and integrated observer is width aw and centre B. Considering that the human spine can
43 shown in Section III. Some contrast experiments and move in a partially restricted manner, thus it is regarded as a
simulation is given in Section IV to test the proposed control constrained branch of the present mechanical system,
44
scheme. Finally, a conclusion of this work is drawn in Section consisting of two upper and lower universal hinges and a
45
V. prismatic pair.
46
47 II. Kinemics and Dynamics With O as the origin, the base coordinate system {B} O-
48 A. Analysis of Physic Structure and Degree of Freedom XYZ is established on the fixed platform, with the x-axis
49 Physic structure of the new back-support exoskeleton is parallel to A3A4 pointing to the A2A4 axis and the y-axis parallel
50 given in Figure 1, where it can be seen that the major structure to AlA4 pointing to A3A4, the z-axis direction can be determined
51 is composed of two supporting platforms (lower chest and according to the right-handed spiral rule. With o as the origin,
52 pelvic supporting platforms) connected by a lumbar curve a dynamic coordinate system {m} o-xyz is established on the
adaptation device and four electric push rods. Both the upper dynamic platform, with the x-axis parallel to the a3a4 pointing
53
platform (lower chest supporting platform) and down platform to the a2a3 axis, the y-axis parallel to the a1a4 pointing to the
54
(pelvic supporting platform) are assembled with a manual a3a4, and the z-axis can be determined according to the right-
55
fastening kit to manually pre-adjust tension on the belt. hand spiral rule.
56 The universal hinge in a branch chain is analysed as 2
Meanwhile, the pressure sensors for the lower chest and pelvis
57 mutually perpendicular rotational hinges (R hinges), where the
are equipped at the inside on platforms to monitor the pressure
58
59
60
Page 3 of 11 IEEE Transactions on Mechatronics

1 R hinge connecting the platform is called the outer R hinge and possible to achieve 5 independent DOF: 2 DOFs of movement
2 the R hinge connecting the branch chain is called the inner R perpendicular to Fp and 3 DOFs of rotation.
3 hinge. The inner R hinge is always perpendicular to the outer B. Analysis of kinematics
4 R hinge and at the same time perpendicular to each branch Noting that in this paper, posture analysis order of this
5 chain li, and the two inner R hinges on each branch chain are mechanical system is z-y-x with Euler angles α, β and γ. The
6 parallel to each other. Let the jth bottom-up R hinge in the ith coordinates of the position of Ai (i=1, 2, 3, 4), ai, o in the base
7 branch chain be named Rij, and then the geometric constraints coordinate system {B} are defined as AiB , aiB and o B ,
8 on the branch chains of the mechanism can be expressed by the
9 following equation: respectively. The coordinates of the position of ai in the
10 Ri1  Ri 2 , Ri 2  li , Ri 2   R03 , R03  R04 (1) dynamic coordinate system {m} are defined as aim and the
11 Therefore, Ri1, R04 and li are co-planar. In the four outer coordinates of ai in the system coordinate system {B} that aiB
12 branches l1, l2, l3, l4, the outer R hinges parallel to the diagonal can be obtained by multiplying the rotation matrix RT with it.
13 of the base platform. In the intermediate branches, R01 is The expression is shown as follows:
14 parallel to the x-axis pointing towards A2A3, R04 is
15  X Ai   xai   X ai   Xo 
perpendicular to the moving platform and the geometric
16 AiB   YAi  , aim   yai  , aiB   Yai  , o B   Yo  , aiB  RT aim  o B , (6)
constraints of the branches can be expressed by the following        
17 equation:  Z Ai   zai   Z ai   Zo 
18 R11   R31   A2 A4 , R21   R41   R21   A1 A3 , R01   A1 A2 , R04  m (2) where
19 From the principle of virtual work, it is clear that the work  xl yl zl  C  C  C  S   S  S  C C  S   C  S  S 
20 done on the joints by the restraining force/moment on each RT   xm ym zm    S  C  S  S   S  C  C S  S   C  S  S  , (7)
21 branch chain of the mechanism is 0. Let the restraining force    
22  xn yn zn   S  C   S C  C 
on each branch chain be Fli, then Fli∙vli = 0 (vli is the linear
23 velocity of ith follower rod), from which it follows that Fli⊥P with S.=sin() and C.=cos(). Then the detailed coordinates of
24 (P is the follower rod on the ith branch chain). Let ωR be the AiB and aim can be obtained by
25 angular velocity of the rotating hinge R in each branch chain   al   al   al   al 
26 of the mechanism, and the restraining moment in the branch  B 1  B 1  B 1  B 1 
A   a , A   a , A
 1 2  w  2 2  w  3 2  w  4 2  aw   a , A 
27 chain is Mli, then Mli ∙ωR = 0, so Mli⊥ωR and (di*Fli) ∙ωR = 0
28   0   0   0   0  (8)
(di is the force arm from the point of connection between the  .
29 rotating joint and the member to Fli). Fli is coplanar with ωR,   bl   bl   bl   bl 
30 namely, coplanar with R. a1m  1  bw  , a2m  1  bw  , a3m  1 bw  , a4m  1  bw 
31  2  2  2  2 
From (2) it is clear that R01 and R04 are not parallel, then   0   0   0   0 
32 combined with (1) it is clear that R01 and R04 are coplanar and
33 According to (6), aiB can be expressed by
therefore they must intersect at a point c. Based on the above
34 described restraint force/moment analysis it is easy to see that   al xl  a2 yl  2 X o   al xl  a2 yl  2 X o 
 B 1  , a B  1 a x  a y  2Y  ,
35 the four outer branches of the mechanism are not subject to  a   a x  a y  2Y
2  2
1 l m 2 m o 2 l m 2 m o

36 restraint force/moment and that there is a restraint force Fp in   al xn  a2 yn  2 Z o   al xn  a2 yn  2 Z o  (9)


37 the intermediate branches that is parallel to R02 past the 
  al xl  a2 yl  2 X o   al xl  a2 yl  2 X o 
38 intersection of R01 and R04. From the fact that the binding force a B  1 a x  a y  2Y  , a B  1  a x  a y  2Y  ,
39 does zero work on the end of the mechanism it follows that  3 2 l m 2 m o 4
2
l m 2 m o

  al xn  a2 yn  2 Z o   al xn  a2 yn  2 Z o 
40 f  v  (d  f )    0, d  c  o (3)
41 Then the length changing of four push rods li (i=1, 2, 3, 4)
where f is directional vector of the constraining force Fp, d is
can be expressed by
42 radius vector from c to o, then in matrix form, (3) can be
43 rewritten by li  aiB  AiB  ( X ai  X Ai )2  (Yai  YAi )2  ( Z ai  Z Ai )2, (10)
44  f T (d  f )T   V  0,V  v   . and the length changing of constraint chain l0 (spine) can be
T
(4)
45 obtained by
46 The degree of freedom (DOF) Ω of this system can be
computed by modified Kutzbach-Grübler theory [1]: l0  o B  X o 2  Yo 2  Zo 2, (11)
47 g
Let the unit vector of each branch chain li be δi and the
48   d (n  g  1)   fi     (5)
49 i 1 vector of the line oaiB be denoted by ei. The expressions for
50 where d=6 is the order of the mechanical system; n=12 is the δi and ei are
51 number of links including frame; g=15 is the number of motion  X ai  X Bi   X ai  X o 
52 units; fi denotes the number of DOF of the ith motion unit; δ is
 i   Yai  YBi  , ei  ai  o   Yai  Yo 
1  B B
(12)
53 the number of redundant constraints in the mechanism; ζ is the li
54 number of local degrees of freedom of the mechanism; There  Z ai  Z Bi   Z ai  Z o 
55 are no redundant constraints and local degrees of freedom in According to the configuration of each rotating hinges in
56 this mechanism, therefore, the DOF of this exoskeleton can be the mechanism, the unit vector of each rotating hinges can be
57 computed by   6  (12  15  1)  (6  2  5  1  4  3)  0  0  5 . written by:
58 According to the constraint forces on the mechanism, it is
59
60
IEEE Transactions on Mechatronics Page 4 of 11

1 Before the analysis of acceleration, it is defined a new


R01  [1 0 0]T , R04  RT  [1 0 0]T  [ zl zm zn ]T (13)
2 symbolic function as follows:
3 From the fact that R01, R04 and l0 are coplanar, it follows
R01  ( R04  Oo)  0 (14)  ux   vx   0 u z u y 
4  
   
u  u y , v  v y , Γ (u )   u z 0 ux   u  v  Γ (u)  v. (23)
5 Looking back to (6) and substituting (13) into (14) can    
6 obtain that  uz   vz   u y u x 0 
 
7 z
Yo  m Z o (15) Let A be the generalised acceleration of centre o of the
8 zn moving platform, a and ε be the linear and angular
9 It is clear that the position and attitude of the parallel accelerations of centre o, respectively, and aL be the input
10 mechanism can be expressed by α, β, γ, Xo, Zo. Let V be the acceleration along the branch chain, with the respective
11 generalized velocity at point o the centre of the moving expressions as follows:
12 platform, v and ω be the linear and angular velocities at point  al 0 
13 o, respectively, and vi be the velocity of the moving platform a 
14 m at the spherical hinge ai; vl is defined as the input velocity of  ax   x   l1 
a 
15 the branch chain. The respective expressions are shown below: A    , a  a y  ,    y  , aL   al 2  . (24)
       
16   vx  x   az   z  a
 l3 
17  v      al 4 
18 V    , v  v y  ,    y  ,
  61  vz  z  Differentiating (20) with respect to time gives the
19 
v  v    e , following acceleration:
20
21  i i
a  (v    ei )  (v    ei ) v2
vl   vl 0 vl1 vl 2 vl 3 vl 4 T , (16) ali   iT eiT         (  ei )   i  li . (25)
22   li li
23 Let the unit vectors of the x, y and z axes of the moving Rewrite the last three items of (25) by
24 platform coordinate system in the fixed coordinate system {B} (v    ei )  (v    ei )  (v  ei   )  (v  ei   )
25 be uo, vo, wo, they can be denoted by
 (v  Γ (ei )   )  (v  Γ (ei )   )   E33  Γ (ei )  V   E33  Γ (ei )  V 
T
26 uo  R  1 0 0T   xl xm xn T (26)
27 
  E33  Γ (ei )
vo  R  0 1 0   yl ym yn  .
T T
28 (17) VT  2
V,
  Γ (ei ) Γ (ei ) 
 wo  R  0 0 1   zl zm zn 
29 T T

30   (  ei )   i  (  ei )  ( i   )
31 Thus ω can be expressed by   Γ (ei )  Γ ( i )   T Γ ( ei )  Γ ( i ) (27)
32   x uo   y vo  z wo (18)
033 033 
33 where ωx, ωy and ωz can obtained by the inverse of (18): VT 
Γ (ei ) Γ ( i ) 
V,
34   yl z l  033
35  (vo  wo )   1  and
 
 x (v  w )  u  y z  
36  m m  i  T
 o o o
 yn zn  vli2  V T    i ei   i V .
T T
(28)
37   
 i i
e
38   xl zl 
 (uo  wo )   1  Through substituting (26), (27) and (28) into (25) , the
39
 y   xm zm    , (19) acceleration mathematical model can be built as
40 ( u  w )  v  
 o o o
 xn zn  ali  J a A  V T H iV (29)
41 
42   xl yl  where Ja is the acceleration Jacobian matrix and
43  (uo  vo )   1   1   Γ (ei )
2
Γ 2 ( i ) Γ (ei ) 
z  (u  v )  w   xm ym    Hi   2  . (30)
44  li -Γ ( i ) Γ (ei ) li Γ ( i ) Γ (ei )  Γ (ei ) Γ ( i ) Γ (ei )66
2

o o o
 xn yn 
45
46 where =xlylzl+xmymzm+xnynzn. Then based on (16), the linear C. Analysis of dynamics
47 velocity of each branch chains li can be represented by To solve for the member forces and moments, it is
48 vli  vi   i  (v    ei )   i   i  v  ei   i   necessary to first calculate the velocities of the individual struts
49 of the system. Each branch chain is divided into two parts: the
 1 2 3 4  (20) part of the chain connected to the base platform is called the
50  0  V .
51 013 e1  1 e2   2 e3   3 e4   4  active rod with lBi denotes the distance from the centre of mass
Combining (4) and (20) gives of the active rod to its connection point Ai within the fixed
52
platform and vBi, aBi are the velocity and acceleration of the
53 vli 0  J v  V (21)
centre of mass of the active rod respectively; The part
54 where connected by the branch chain and the moving platform is
55  1 2 3 4 f  (22) called the follower rod (telescopic rod). lmi is used to represent
Jv   0 
56 013 e1  1 e2   2 e3   3 e4   4 d  f 66 the distance between the center of mass of the telescopic rod
57 is the velocity Jacobian matrix. and the connection point ai of the moving platform. vmi and ami
58
59
60
Page 5 of 11 IEEE Transactions on Mechatronics

1 are the speed and acceleration of the center of mass of the


2 vmi  (lmi  li ) Γ (i ) J i  iiT iiT Γ (ei ) 
follower rod, respectively. Since the driver and follower are 
  mi mi 
J V , J   (44)
3 collinear, their angular velocities are the same as follows:  li   J i 
4 R 
li  i1 Ri1  i 2 Ri 2 , Ri 2  i1 i , (31) For each branch chain, the linear acceleration of the
5
Ri1   i centre of mass of its telescopic rod can be represented by
6 aBi   li  i (li  lmi )  li  (li  i )(li  lmi )  i ali  2(li  i )vli . (45)
7 where θ̇ i1 and θ̇ i2 are angular velocity of Ri1 and Ri2,
respectively. Using Aiai to cross the two sides of equation (31) It is defined that the external force and torque on the
8
gives that center of mass of the moving platform is fp and np, respectively;
9
Im
o denotes the inertia matrix of the centre of mass of the
10 li  Ai ai  i1 Ri1  Ai ai  i 2 Ri 2  Ai ai
moving platform with respect to the moving coordinate system;
11  vi  vli i  vi  vl  i   i (32) MM is the mass of moving platform; fL and TL are the load (in
12 this paper, it means the upper body). Then based on the above
 ( I 33   i   )vi   Γ ( i ) v  Γ ( i ) Γ (ei ).
T 2 2
13 i
definition, the forces and moments acting on the centre of mass
14 Then multiplying (32) by Ri2 and Ri1 can respectively
of the moving platform can be represented by
15 leads to
 fp   f L  M M g  ma a 
16 i1 ( Ri1  Aai i ) Ri 2  (vi  vli i ) Ri 2  (  Γ ( i ) v  Γ ( i ) Γ ( ei ) ) Ri 2
2 2
(33) Tp      , (46)
n
   L T  RI m T
R     ( RI m T
R  ) 
17 and p o o

18 i 2 ( Ri 2  Ai ai ) Ri1  (vi  vlii ) Ri1  ( Γ (i )2 v  Γ (i )2 Γ (ei ) ) Ri1. (34) where fL=[fLx fLy fLz] and TL=[TLx TLy TLz] are the load and
19 moment of loads on the moving platform. The forces and
Then θ̇ i1 and θ̇ i2 can be obtained by moments acting on the centre of mass of the active rod and the
20
21 ( Γ (i )2 v  Γ (i )2 Γ (ei ) ) Ri 2 ( Γ (i )2 v  Γ (i )2 Γ (ei ) ) Ri 2 centre of mass of the telescopic rod are
 i1   (35)
22 ( Ri1  li ) Ri 2 ( Ri1  Ri 2 )li  f Bi   mBi g  mBi aBi 
TBi      B i  (47)
23 and 
 Bi   i mi i
n R I ( R B T
)  Bi  mi  ( R B i
I
i mi ( R i
B T
) i 
)
24 ( Γ (i )2 v  Γ (i )2 Γ (ei ) ) Ri1 ( Γ ( i )2 v  Γ ( i )2 Γ (ei ) ) Ri1 and
25  i1   . (36)
( Ri 2  li ) Ri1 ( Ri1  Ri 2 )li  f mi   mmi g  mBi aBi 
26 Tmi      B i  . (48)
Combining (31), (35) and (36) yields  nmi    Ri I mi ( Ri )  mi  mi  ( Ri I mi ( Ri ) mi ) 
B T B i B T
27
( R  R )  (v  vli i ) where IiBi is the inertia matrix of the driving rod at its center
28 li  i1 Ri1  i 2 Ri 2  i1 i 2  J iV (37)
29 li i  ( Ri1  Ri 2 ) of mass; Iimi is the inertial matrix of the telescopic rod at its
30 and center of mass; can mBi and mmi are the mass of the active
31  ( R RT  Ri 2 RiT1 ) Γ ( i )2 ( Ri1 RiT2  Ri 2 RiT1 ) Γ ( i )2 Γ ( ei )  rod and telescopic rod, respectively. RBi is the rotation matrix
32 J i   i1 i 2  . (38) can be represented by
 ( Ri1  Ri 2 )li ( Ri1  Ri 2 )li 
33 RiB   Ri 2  i  Ri 2  i  . (49)
Taking the derivative of (37) with respect to time gives
34 According to the principle of virtual work, the balance
the following angular acceleration εli of li:
35 equation of drive forces τp can be represented by
36 ( R ( R  R ))  (v  v  )  ( R  R )  (a  a   v    )
 li  i1 i1 i1 i 2 i li i i1 i 2 i li i li li i  v 4
 v  v 
37 lii  ( Ri1  Ri 2 )  p vl  Tp      TBi  Bi   Tmi  mi    0. (50)
w    i 0      
38 ( Ri1  Ri 2 ))  (vi  vlii )  i1lii Ri 2  (vlii  lili  i )  ( Ri1  Ri 2 )
Bi mi

39 (39) By combining (22), (41), (44) and (50) obtains


40 lii  (Ri1  Ri2 )2 4
 p JV  TpV   (TBi J BiV  Tmi J miV )  0 (51)
41 ( R  R )  (a  a   v    )   R  ( v  v  )  (v   l    )  ( R  R )   i 0
42  i1 i 2 i li i li li i i1 i 2 i li i li i i li i i1 i 2 li .
lii  ( Ri1  Ri 2 ) or
43 4

44
For each branch chain, the velocity of the centre of mass  p   J 1TpV  J 1  (TBi J Bi  Tmi J mi ). (52)
of its active rod is i 0
45
vBi  li  i lBi  lBi Γ (i )li  lBi Γ (i ) J liV (40) Through taking (46), (47) and (48) into (52) can give
46 f    M g   m g   m g   
Retrospecting (37) and (38) leads to (53)
4
 p   J 1  L   J 1   M     J Bi  Bi   J Bi  mi    
47 TL    0  i 0   0   0   
48 vBi   lBi Γ ( i ) J i 
   J BiV , J Bi  
  M g  4    
. (41)  m a   m a
J 1   mM T     J Bi  B i Bi BBi T   J mi  B i mi Bmi T   
49  li   J i    RI o R   i 0   Ri I Bi ( Ri )  Bi   Ri I mi ( Ri )  mi   

50 For each branch chain, the linear acceleration of the   0  4   0   0   


J 1       J Bi    R B I i ( R B )T     J mi    R B I i ( R B )T      .
51    ( RI o R  )  i 0 
m T
 i  i Bi i i 
  i  i mi i i 
  
centre of mass of its active rod can be represented by
52 aBi   li  i lBi  li  (li  i )lBi . (42) (53) gives the driving force required for the support rods
53 of the exoskeleton to move the moving platform to a certain
The telescopic rod centre of mass velocity is
54 posture, thereby regulating the power input to the exoskeleton.
vmi  li   i (li  lmi )   i vli  (li  lmi ) Γ ( i )li   i  iT eiT   iT V
55 III. Control system
(43)
56
57

 (lmi  li ) Γ (i ) J i   i iT  i iT Γ (ei ) V .  Then for better establishing control scheme, (53) can be
rewritten by a close dynamics equation [32]. Considering the
58 Retrospecting (37) and (38) again leads to
59
60
IEEE Transactions on Mechatronics Page 6 of 11

1 delay process brought from the buffer, the dynamic model of


2 e L  x1  z1
the exoskeleton can be rewritten by z  z  l e
3 1
M ( x ) x  C ( x, x ) x  G  τ (t   du (t ))+Fd , (54) 
2 1 L
, (58)
4  2z  B ( x ) τ  z3  l 2 e L
5 or
 z3  l3e L
6 x   M 1 ( x )C ( x, x ) x  M 1 ( x )G + M 1 ( x )Fd + M 1 ( x )τ (t   du (t )), (55)
7 where x=[Xo, Yo, Zo, α, β, γ] is the position and attitude where z1, z2 and z3 are estimates of x1, x2 and x3, respectively.
8 information of the moving platform in the based frame, 𝜏du≥0 With proper selection of the parameter l1, l2, l3, the observer
state variable z1, z2 and z3 can track x1, x2 and x3 with favorable
9 is input delay of the control torque. M(x) is the inertia matrix
performance. As a relatively mature technique, the detailed
10 and C(x,𝑥̇ ) is the Coriolis and centripetal terms. It is assumed
analysis of LESO can refer to some literature [29], [30].
11 that the measurement process of state variables is also with
Considering the LESO constrained by its bandwidth, a SMO is
12 time-varying delay 𝜏ds ≥ 0. Due to the buffers and sensors
then defined to compensate the tracking error of LESO. Define
13 equipped on the each supporting rods are in the same type, it the tracking error of LESO as
14 is reasonable to assume that each state variable and control
Fe  F ( x, x )  z3 . (59)
15 torque is subject to same delay 𝜏ds and 𝜏du, respectively. For
16 better analysis, xij(t) i=1,…,6 is chosen to denote the full Assumption 1: F( 𝑥̇ ,x) and its derivative are bound,
17 system state variables, where i is the ith elements in x and j is meanwhile, ||Fe||and its derivative ||𝑭̇𝑒 || are bounded by ε1
18 its order. Use τi i=1,…,6 to denote the control torque of six and ε2, respectively.
DOF corresponding to x=[Xo, Yo, Zo, α, β, γ]. For simplification, Then define a sliding mode surface as follows:
19
20 (55) can be rewritten by s  z2   (60)
21 x  B( x )  τ (t   du (t ))  F ( x, x ), (56) where  satisfies
22 where 𝑭(𝒙,̇ 𝒙) is the lumped disturbance including the   B( x )τ  Fˆe  z3  1 sgn( s)  l2 eL (61)
23 external disturbance, measurement error and modeling errors, Hence, a SMO can be built up by
24 etc., B(x)=M -1(x) is the control matrix.
Assumption 1: The delay of state variable 𝜏ds is considered  Fˆe  J  2 z2

25  (62)
ˆ
26 exactly known and 𝜏du is with a known upper bound τ̅du 0,  J  2 ( z3  B( x )τ  Fe  l2 eL )  3 sgn( s)

27 meanwhile, the differentials of 𝜏ds and 𝜏du are bounded by 1. where Fˆe is estimate of tracking error of LESO, λ1, λ2 and λ3 are
28 There is existing a total upper bound τ̅𝑑𝑡 (t)≥τds (t)+τ̅du (t) for
the total delay process including state variable and control adjustable gains. Recalling eL=x - z1 and looking back to LESO
29 in (58) can give
30 torque. The function of delay is changing slowing, thus 1 >
τ̅̇du > 𝜏̇ 𝑑𝑢 , and 1 > τ̅̇ds > 𝜏̇ 𝑑𝑠 . s  z2    Fe  1 sgn( s) (63)
31
32 Assumption 2: The reference trajectory yd=[Xd, Yd, Zd, αd, where Fe  Fe  Fe . ˆ
33 βd, γd] is are defined on [−τ̅𝑑𝑡 ,+∞), are known at each time and Theorem 1: For the SMO (62), if the parameters satisfy
34 Lipschitz. that
35 Assumption 3: It is assumed that the control torque does
not work during the delay domain, namely, 1  1  Fˆe  a
36 
37 τ  0 t [ dt   du ,  du ) . 2  0

38 Assumption 4: The desired trajectory yd,i and its derivative 3   2 (64)
39 𝑦̇ 𝑑,𝑖 of the exoskeleton is bounded. where a 0 is a sharping gain with small positive value, then
40 A. Integrated observer the estimate error Fe of SMO will converge into bounded set in
41 This part is devoted to establishing a linear extended state
42 finite time.
observer (LESO) to estimate the external disturbance and Proof: Through combining (56) and (62) can obtain that
43 internal uncertainties, additionally, a sliding mode observer
44 (SMO) is design to compensate the remaining tracking errors Fe  Fe  2 Fe  3 sgn( s) (65)
45 of LESO so that enhance its convergence rate. Define a following Lyapunov function:
46 In sate space, the dynamic model of BSE can be 1
47 expressed as Vs1  sT s (66)
2
48  x1  x2 Then taking the time derivative of Vs1 yields that
49  Vs1  sT s  sT ( Fe  1 sgn( s ))
 x 2  B ( x ) τ  x3 , (57)
50   Fe  s  (1  Fˆe  a ) s  (1  Fˆe  Fe  a ) s (67)
51  x3  F ( x , x )
52 Where x1=x, x2=𝒙̇ and x3=F(𝑥̇ ,x) is the system state variable  (1  Fˆe  a  Fe  Fˆe ) s
53 and lumped disturbance. Based on (57), the following LESO =  (1  Fe  a ) s  0
54 can be built up: It thus follows that the slip surface s converges to zero in
55 finite time. Furthermore, its rate of convergence is related to λ1.
56 Define the following Lyapunov function by the tracking
57 error of SMO:
58
59
60
Page 7 of 11 IEEE Transactions on Mechatronics

1 where
1 T
2 Vs 2 
Fe Fe (68)
3 2  i,1  0
Taking the time derivative of Vo2 yields 
4  i ,1  0
5 Vs 2  FeT ( Fe  2 Fe  3 sgn( s))   
(69)  0 xi ,1 ( s0 )  ci ,0 ( s0 )  (e i ,1 dt  1) i,1
6  2 FeT  Fe  Fe  Fe  3 FeT  i ,1  . (79)
 
7  e i ,1 dt
Recalling (64) gives
8 Subsequently, the intermediate control signals for first
9 Vs 2  2 Fe T  Fe  Fe  Fe  3 Fe order differential process can be given by
(70)
10  2 Fe T  F  Fe Fe   2 Fe  0  x ( 0 )  ci ,0 ( s0 ) 
11 ci ,1 ( xi ,1 , t )  ki ,1  Q  i ,1 s  . (80)
 i ,1 ( s )
12 Therefore, the tracking error Fe will converge into bounded  
13 set in finite time. where ki,0 0(i=1,…,6) is the control gain.
14 B. prescribed performance controller Then for the second order differential process of this
15 Through using the output of integrated observer to design system, select the performance functions as
16 feedforward compensation can make the dynamic model i ,2 (t )  ( i0,2  i,2 )e i ,2 t  i,2 , (81)
17 transfer into where
18 x  B( x )  τ (t   du (t ))  Fe , (71)  i,2  0
19 where Fe is the tracking error of the integrated observer for 
20  i ,2  0
lumped disturbance and influenced by the performance of 
21  0
 
xi ,2 ( s0 )  ci ,1 ( s0 )  (e i ,2 dt  1) i,2
integrated observer.
22 
 i ,2    . (82)
Define the delay performance function of state variable  e i ,2 dt
23
as Eventually, the corresponding control torque can be built
24
 s (t )  t   ds (72) up by
25
Recalling τ̅𝑑𝑡 (t)≥τds (t)+τ̅du (t) and Assumption 3 gives 2ki ,2Q[ i (t )]
26
that ζs (t) ≥ −τ̅𝑑𝑡 and ζ𝑠 (t) − τ̅𝑑𝑢 ≥ −τ̅𝑑𝑡 − τ̅𝑑𝑢 in
 i   sgn  B( x)  , B( x)  B( x)  BT ( x) (83)
27 i ,2 ( s )[1   i (t )]
0 0
28 [−τ̅𝑑𝑡 ,+∞). Moreover, it is defined that ζs =ζs ( − τ̅du ). Then an where
29 exponentially decaying output performance function is defined
xi ,2 ( s )  sgn  B( x )   
t

30 for as follows:  ( s)ds  ci ,1


 s  du i
 i (t )  , (84)
31 i ,0 (t )  ( i0,0  i,0 )ei ,0 t  i,0 . (73) i ,2 ( s )
32 where
33 and sgn(A)=1 means A is positive define and sgn(A)=−1 means
 i,0  0 A is negative define.
34  The performance attributes required for the output
35  i ,0  0 tracking error can be defined directly by an appropriate choice
36   
 0 xi ,0 ( s0 )  yd ,i ( s0 )  (e i ,0 dt  1) i,0 of ρi,0 in equation (73). Therefore, the goal ρi,0 of is to constrain
37  i ,0  . (74) the tracking errors, namely,
 
38  e i ,0 dt
In (74), ρ∞i,0 is set to constrain the tracking error at xi ,0  yd ,i  i ,0 , t  0, i  1,...,6, (85)
39
40 steady state and δi, i=1,…,6 is the shaping gain can modify the Let’s define a following expression ξij for i=1,…,6
41 convergence rate of the tracking error. x ( )  yd ,i ( s )
i ,0  i ,0 s , (86)
42 Define a function as follows: i ,0 ( s )
43 1 x 
Q ( x )  ln   (75) xi ,1 ( s )  ci ,0
44 1 x  i ,1  , (87)
45 i ,1 ( s )
It is clear that in x=[-1, 1], Q(x) is strictly increasing and
46 t
satisfies that xi ,2 ( s )  sgn[ B( x )]    i ( s)ds  ci ,1
47 i ,2 
 s  du
 lim Q ( x )   . (88)
48 x 1 i ,2 ( s )
49  . (76)
 xlim Q ( x )   Define the error transformation as
1
50 eij  eijQ(ij ), i  1,...,6, j  0,1; (89)
51 Then an initial intermediate control signals can be built
52 by T
en  e1,n , , e6,n  , n  0,1, 2 (90)
53  
ci ,0 ( xi ,0 , t )  ki ,0  Q  xi ,0 ( s )  yd ,i ( s )  / i ,0 ( s ) . (77) where eij is the upper bound of eij.
54 where ki,0 0(i=1,…,6) is the control gain. Then it can be obtained that
55 Then for the first order differential process of this system, c j  k j e j , j  0,1;, (91)
56 select the performance functions as
57 τ = sgn[ B( x)]k2 q2 e2 , (92)
i ,1 (t )  ( i0,1  i,1 )e i ,1 t  i,1 , (78)
58
59
60
IEEE Transactions on Mechatronics Page 8 of 11

1 where can ensure ki,0 ∙|ei,0 |≥λ̅i,0 to make 𝑉̇0 ≤ 0 , so that the error
2
k j  diag[k1, j ,...k6, j ], k2  diag[k1,2 ,...k6,2 ], (93) can converge into a bounded range.
3
q2  diag ( q1,2 ,..., q6,2 ), (94) For the first order process, select a new Lyapunov
4
function by the second error vector as follows:
5 dQ (i , j ) 2 i, 1j ( s )
qi , j  i, 1j ( s )  , j  0,1, 2., (95) 1
6 V1  e1T e1 . (110)
d i , j 1  i2, j 2
7
8 Taking time derivative of (86)-(88) yields Then repeat the proof steps (103)-(109) of the first
9 
ξi ,0 =i,01 (1   ds ) ρi ,1ξi ,1  (1   ds )  ρi ,1ξi ,1  yd ,i  ki ,0 ei ,0   (96)
Lyapunov function can obtain that there is always existing ki,1
can ensure 𝑉1̇ ≤ 0.
10
11 ξi ,1 =ρi,11  (1   ds ) ρi ,2 ξi ,2  (1   ds )  ρi ,2 ξi ,2  ci ,0  ki ,1ei ,1   (97) For the second order process, select another Lyapunov
function by the second error vector as follows:
12
13 (1   ds ) Fe  B( x )τ  ρi,21ξi ,2  sgn[ B( x)]τ   ...  1
V2  e2T kn e2 . (111)
1
ξi ,2  ρ   (98) 2
14 i ,2
...  (1   ds ) ρi ,1 ξi ,1  ci ,1  ki ,2 qi ,2 ei ,2
1

15  Taking the time derivative of V2 and according to (98), it


where can be obtained that
16
17 ρi , j  diag ( 1, j ,..., 6, j ) (99) V2  (1   ds )e2T ki ,2 q2 sgn[ B( x )]( B( x)  B( x) I )ki ,2 q2 e2
(112)
18 τ   1 (t   du (t )),..., 6 (t   du (t )) (100) (1   ds )e2T ki ,2 q2  Fe  ρi ,2 ξi ,2  c1  sgn[ B( x)]τ  ,
19
20 τ   1 ( s   du (t )),..., 6 ( s   du (t )) (101) where I∈R6×6 is an unit matrix. It is defined that
21 Theorem 1: For system (56) with control input and state P1  sgn[ B( x)]( B( x)  B( x) I ), (113)
22 measurement delays 𝜏ds ≥ 0 and 𝜏ds ≥ 0, respectively, the P2  Fe  ρi ,2 ξi ,2  c1  sgn[ B( x )]τ . (114)
23 prescribed performance controller (73)-(84) can drive the ̇
Then V2 comes to
24 output of the system track the desired trajectory yd,i with
25 bounded errors. V2  (1   ds )  e2T ki ,2 q2 P1ki ,2 q2 e2  e2T ki ,2 q2 P2  . (115)
26 Proof of Theorem 1: Define the first Lyapunov function Owing to the disturbance F(x) is unknown but naturally
27 by the first error vector as follows: bounded, as well as, 𝑩 ̅ (x) and B(x) derive from the inertial
28 1
V0  e0T e0 (102) matrix M(x) thus they are bounded too, therefore, P1 is
29 2 bounded. According to the definition in (91), namely,
30 Taking the derivative of V0 with respect to time and c1=˗ki,1∙ei,1, thus is bounded. ρ̇ 2 is bounded and the is
31 looking back to (96) yields conclusion can be quoted from (104). Eventually, it can be
32 V0  e0T q0 [(1   ds ) ρi ,1ξi ,1  (1   ds )( ρi ,0 ξi ,0  yd ,i  ki ,0 e0 ] inferenced that P1 is bounded. Then we have
33 (103)
 e0T q0 (1   ds )ki ,0 e0  e0T q0 (1   ds )( ρi ,1ξi ,1  ρi ,0 ξi ,0  yd ,i ) 
 
6

34 V2   (1   ds ) ei,2
2
ki,2 qi ,2 Pi ,1ki,2 q2  ei 2 ki,2 qi ,2 Pi ,2
 e0T q0 (1   ds )  ki ,0 e0  ( ρi ,1ξi ,1  ρi ,0 ξi ,0  yd ,i )  i 1
35 (116)
Looking back to (73), (86) and (87) gives 6
36  t
  (1   ds ) ei 2 ki,2 qi ,2  Pi ,1ki,2 q2  Pi ,2 
37 ρi ,0 ( s )  ( ρi,0  ρi0,0 )(1   ds ) i ,0 e i ,0   i ,0 (1   ds )( ρi ,0  ρi,0 ). (104) i 1

38 Therefore, 𝜌̇ 0 is bounded. It is clear that 1 > 1 − 𝜏̇ 𝑑𝑠 > Therefore, there is always ki,2 available to make
39 0 in accordance with assumption 1, thus it is bounded, ρ1 ,ρ2 Pi ,1ki,2 q2  Pi ,2  0, (117)
40 are bounded in accordance with its defined characteristic, 𝑦̇ 𝑑𝑖 So that ensure V̇ 2 ≤0, then ensure ei,2 can converge into a
41 is assumed bounded and ξi,0 and ξi,1 are compacted in the bounded set in finite time domain.
42 bounded range [-1, 1], which lead to that IV. Simulation and experiments
43 ki ,0 e0  ( ρi ,1ξi ,1  ρi ,0 ξi ,0  yd ,i )  λ0 (105) This section is devoted to testing the performance of the
44 ̅ ̇
is bounded, namely, |𝜆0 | < 𝜆0 . Then 𝑉0 can be expressed by proposed controller. The main structural parameters of the BSE
45
V0  e0T q0 (1   ds )( ki ,0 e0  λ0 ). (106) are shown in Table I and The experimental equipment and
46 subject are shown in Figure 3. Given the five independent
47 Therefore, with appropriately selected control gain ki,0
positional output parameters A of the moving platform, the
48 can ensure ki,0 ∙|e0 |≥λ̅i,0 to make 𝑉̇0 ≤ 0, so that the error can positional parameter Yo can be found according to equation
49 converge into a bounded range. (15). Set the motion law of the moving platform as shown in
50 It is defined that equation (118) and its diagram is shown in Figure 4 and Figure
51 λ0 =( ρi ,1ξi ,1  ρi ,0 ξi ,0  yd ,i ) (107) 5.
52 Then we have  X o   30cos(0.2 t   ) 
53 V0  e0T q0 (1   ds )ki ,0 e0  e0T q0 (1   ds ) λ0 (108)  Z  300  30cos(0.2 t   ) 
54  o  
x  30cos(0.2 t )  (118)
 
6
55 V0   qi ,0 (1   ds ) ei ,0 i.0  ki ,0 ei ,0 (109)    
56 i 1
    30cos(0.2 t   ) 
57 Therefore, with appropriately selected control gain ki,0     15cos(0.2 t   ) 
58
59
60
Page 9 of 11 IEEE Transactions on Mechatronics

1 In this paper, the non-singular terminal sliding mode overshoot than the other two controllers, which largely reduces
2 controller (NTSMC) and PID proposed by [31] are used to the discomfort to the wearer from sudden surges in position
3 compare the proposed control scheme. The two control law are change.
4 shown as follows:
5 t
6 τ pid  K p e  K i  edt + K d e, (119)
0
7 g
8 τ NTSMC  xd  k1 sgn(e )2h / g  k2 sgn( s), (120)
h
9 where Kp, Ki, and Kd are matrices of control gain of PID
10 controller, e=x-xd is the trajectory tracking errors,
11 s=e+𝑘1−1 [sgn(𝒆̇ )]𝑔/ℎ . k1, k2, g and h are parameters of NTSMC.
12 Table I physical parameters of the BSE
13 Items Value
14 Mass of the moving platform MM 680 g
Mass of the base platform MB 750 g
15 Generalized length of the moving platform bl 380 mm
16 Generalized width of the moving platform bw 185 mm
Generalized length of the base platform al 350 mm
17 Generalized width of the base platform aw 150 mm
Figure 5 The reference posture of moving platform
18 Maximum length of supporting rods limax 350mm
Minimum length of supporting rods limin 150mm Moving to Figure 7, where the tracking errors of the
19 Moment of inertia of moving platform Im 10-2×diag(2.089, 2.089, 4.178) kg∙mm2
Mass of the telescopic rods MT 90g proposed integrated observer is given, we can found that the
20 Mass of the inertia of active rods MA 120g observer can track the lumped disturbance with little errors. In
21 Moment of inertia of telescopic rods Iti 10-3×diag(2.273, 2.273, 0.4725) kg∙mm2
the control scheme, integrated observer is key to address the
Moment of inertia of active rods Iai 10-2×diag(2.563,2.563,0.126) kg∙mm2
22 modeling errors, measurement noise and external disturbance,
23 which makes up the forward compensation law to reduce the
24 control burden of the controller.
25
26
27
28
29
30
31
32
33 Figure 3 Experimental equipment and subject
34 Figure 6 Tracking error comparison of the three control scheme
35
36
37
38
39
40
41
42
43 Figure 7 Tracking performance of the integrated observer
44 Finally, the change of length, velocity and acceleration of
45 the supporting rods are shown in Figure 8~Figure 10,
46 respectively. For the set law of motion of the moving platform,
47 there is a small variation in the expansion and contraction of
Figure 4 The reference position of moving platform
48 the five supporting rods, but all meet the design requirements
The simulation results are shown in Figure 6~Figure 10. of the mechanism. The length of the chain l0 (the spine) is
49 In Figure 6, it can be seen that compare to the PID controller
50 always in the middle of the four external chains, which is also
and NSTMC, the proposed controller can drive the moving in accordance with the ergonomics and the geometrical
51 platform following the desired trajectory with less errors. The
52 characteristics of the BSE. As can be seen in Figure 8, the
desired trajectory is from the gyroscope attached on the body, length of the drive chain varies smoothly, which facilitates the
53 which means the compactness and synergy rate between body control of the drive rod. It can be seen in Figure 9 that the
54 and BSE can be reflected by the tracking errors. Supporting the velocity of branch chain l1 and l2 is large, while that of spine l0
55 upper body while driving the upper platform to follow the is relatively small, this can protect the spine from fast fraction.
56 movement of the lumbar spine can relax and relieve the As can be seen from Figure 10, the acceleration of rods l1, l2,
57 patient's lumbar spine and lumbar muscles. In addition, the l3 and l4 is higher, while the acceleration of rod l0 is relatively
58 proposed controller has a faster response time and lower
59
60
IEEE Transactions on Mechatronics Page 10 of 11

1 small. In the selection and design of the parallel mechanism, exoskeleton. This controller can deal with the time delay effect
2 the performance of the driving chains l1, l2, l3 and l4 should be from multi dimensions and be with no need for the complex
3 better than that of the spine l0. In general, the speed and decoupling process. Additionally, a new observer integrated by
4 acceleration change curve has a smooth transition and no linear extended state observer and sliding mode observer is
5 sudden change point, so there is no shock and vibration caused proposed to estimate the lumped disturbance including
6 by it during the movement, which is conducive to ensuring the modeling errors, measurement noise and external disturbance,
7 dynamic performance of the BSE and thus protecting the then the estimates of lumped disturbance are used to design
8 lumbar vertebrae and lumbar muscles. feedforward compensation to simplify the dynamic model of
9 BSE so that enhance the control performance of the proposed
10 prescribed performance controller. This type of control scheme
11 can be extended to other second order control systems with
12 time delay phenomenon. Simulation results and tests verify
13 that the control method proposed in this paper allows the BSE
14 to effectively support and follow the movement of the upper
15 body, and that the new observer also provides a good estimate
16 of the overall disturbance.
17 Acknowledgement
18 This work is supported by Natural Science Foundation of
19 Sichuan (2023NSFSC0029). The authors would like to thank
20 all the anonymous reviewers for their comments on the
21 Figure 8 Change of length of the supporting rods refinement of this paper.
22 Reference
23 [1] Safety Index, Liberty Mutual Insurance Company, Boston, MA, USA,
24 2017.
[2] Alemi M M, Madinei S, Kim S, et al. Effects of two passive back-
25 support exoskeletons on muscle activity, energy expenditure, and
26 subjective assessments during repetitive lifting[J]. Human factors, 2020,
27 62(3): 458-474.
[3] Yandell M B, Wolfe A E, Marino M C, et al. Effect of a Back-Assist
28 Exosuit on Logistics Worker Perceptions, Acceptance, and Muscle
29 Activity[C]//Wearable Robotics: Challenges and Trends: Proceedings
30 of the 5th International Symposium on Wearable Robotics, WeRob2020,
and of WearRAcon Europe 2020, October 13–16, 2020. Springer
31
International Publishing, 2022: 7-11.
32 [4] Lamers E P, Yang A J, Zelik K E. Feasibility of a biomechanically-
33 assistive garment to reduce low back loading during leaning and
Figure 9 Change of velocity of the supporting rods lifting[J]. IEEE Transactions on Biomedical Engineering, 2017, 65(8):
34
1674-1680.
35 [5] De Bock S, Ghillebert J, Govaerts R, et al. Benchmarking occupational
36 exoskeletons: An evidence mapping systematic review[J]. Applied
37 Ergonomics, 2022, 98: 103582.
[6] Theurel J, Desbrosses K. Occupational exoskeletons: overview of their
38 benefits and limitations in preventing work-related musculoskeletal
39 disorders[J]. IISE Transactions on Occupational Ergonomics and
40 Human Factors, 2019, 7(3-4): 264-280.
[7] Zhu Z, Dutta A, Dai F. Exoskeletons for manual material handling–A
41 review and implication for construction applications[J]. Automation in
42 Construction, 2021, 122: 103493.
43 [8] Li X G, Hou C, He J. Saturated Sliding Mode Control Scheme for a
44 New Wearable Back-Support Exoskeleton[J]. IEEE Transactions on
Automation Science and Engineering, 2023.
45 [9] Poliero T, Fanti V, Sposito M, et al. Active and passive back-support
46 Figure 10 Change of acceleration of the supporting rods exoskeletons: a comparison in static and dynamic tasks[C]//2022 9th
47 V. Conclusion IEEE RAS/EMBS International Conference for Biomedical Robotics
and Biomechatronics (BioRob). IEEE, 2022: 01-08.
48 The key to the BSE's ability to assist the upper body is [10] Erezuma U L, Amilibia M Z, Elorza A E, et al. A Statistical Parametric
49 the need to drive the upper platform to track the lumbar spine's Mapping Analysis Approach for the Evaluation of a Passive Back
50 posture quickly and accurately while supporting the body's Support Exoskeleton on Mechanical Loading During a Simulated
weight, but due to the time lag caused by some flexible Patient Transfer Task[J]. Journal of Applied Biomechanics, 2023,
51 1(aop): 1-12.
52 materials, the drive sometimes cannot drive the platform to [11] van Sluijs R M, Wehrli M, Brunner A, et al. Evaluation of the
53 follow the prescribed trajectory quickly enough. The key to the physiological benefits of a passive back-support exoskeleton during
54 BSE's ability to assist the upper body is the need to drive the lifting and working in forward leaning postures[J]. Journal of
upper platform to quickly and accurately track the posture of Biomechanics, 2023, 149: 111489.
55 [12] Alemi M M, Simon A A, Geissinger J, et al. Modeling the metabolic
56 the lumbar spine while supporting the body weight. In this reductions of a passive back-support exoskeleton[J]. Journal of Applied
57 paper, therefore a new prescribed performance controller is Physiology, 2022, 132(3): 737-760.
proposed to deal with the time delay effects of the back-support [13] Schwartz M, Desbrosses K, Theurel J, et al. Using passive or active
58
59
60
Page 11 of 11 IEEE Transactions on Mechatronics

1 back-support exoskeletons during a repetitive lifting task: influence on


2 cardiorespiratory parameters[J]. European Journal of Applied
photos and the bios 1
3 Physiology, 2022, 122(12): 2575-2583.
[14] Kennard M, Yagi K, Hassan M, et al. Variable-Damper control using
4 MR fluid for lower back support exoskeleton[J]. IEEE/ASME
5 Transactions on Mechatronics, 2023, 28(1): 579-587.
6 [15] Lanotte F, McKinney Z, Grazi L, et al. Adaptive control method for
dynamic synchronization of wearable robotic assistance to discrete
7 movements: Validation for use case of lifting tasks[J]. IEEE
8 Transactions on Robotics, 2021, 37(6): 2193-2209.
9 [16] Hyun D J, Lim H S, Park S I, et al. Singular wire-driven series elastic
10 actuation with force control for a waist assistive exoskeleton, H-
WEXv2[J]. IEEE/ASME Transactions on Mechatronics, 2020, 25(2):
11 1026-1035.
12 [17] Sun Y, Tang Y, Zheng J, et al. From sensing to control of lower limb
13 exoskeleton: A systematic review[J]. Annual Reviews in Control, 2022,
53 (2022) 83–96.
14 [18] Tucker M R, Olivier J, Pagel A, et al. Control strategies for active lower
15 extremity prosthetics and orthotics: a review[J]. Journal of photos and the bios 2
16 neuroengineering and rehabilitation, 2015, 12(1): 1-30.
[19] Masengo G, Zhang X, Dong R, et al. Lower limb exoskeleton robot and
17 its cooperative control: A review, trends, and challenges for future
18 research[J]. Frontiers in Neurorobotics, 2023, 16.
19 [20] Baud R, Manzoori A R, Ijspeert A, et al. Review of control strategies
for lower-limb exoskeletons to assist gait[J]. Journal of
20 NeuroEngineering and Rehabilitation, 2021, 18(1): 1-34.
21 [21] Chen Z, Guo Q, Li T, et al. Gait prediction and variable admittance
22 control for lower limb exoskeleton with measurement delay and
23 extended-state-observer[J]. IEEE Transactions on Neural Networks and
Learning Systems, 2022, (Early Access).
24 [22] Kim S, Bae J. Force-mode control of rotary series elastic actuators in a
25 lower extremity exoskeleton using model-inverse time delay control[J].
26 IEEE/ASME Transactions on Mechatronics, 2017, 22(3): 1392-1400.
[23] Ding M, Nagashima M, Cho S G, et al. Control of walking assist
27 exoskeleton with time-delay based on the prediction of plantar force[J]. photos and the bios 3
28 IEEE Access, 2020, 8: 138642-138651.
29 [24] Beik-Mohammadi H, Kerzel M, Pleintinger B, et al. Model mediated
teleoperation with a hand-arm exoskeleton in long time delays using
30 reinforcement learning[C]//2020 29th IEEE International Conference
31 on Robot and Human Interactive Communication (RO-MAN). IEEE,
32 2020: 713-720.
33 [25] Han S, Wang H, Tian Y. Model-free based adaptive nonsingular fast
terminal sliding mode control with time-delay estimation for a 12 DOF
34 multi-functional lower limb exoskeleton[J]. Advances in Engineering
35 Software, 2018, 119: 38-47.
36 [26] Han S, Wang H, Tian Y. A linear discrete-time extended state observer-
based intelligent PD controller for a 12 DOFs lower limb exoskeleton
37 LLE-RePA[J]. Mechanical systems and signal processing, 2020, 138:
38 106547.
39 [27] Fridman E. Introduction to time-delay systems: Analysis and
control[M]. Springer, 2014.
40 [28] Zhang Y, Mu D. New concept and new theory of mobility calculation
41 for multi-loop mechanisms [J]. Science China Technology Science,
42 2010, 53(6): 1598-1604.
[29] Yoo D, Yau S S T, Gao Z. On convergence of the linear extended state
43 observer[C]//2006 IEEE conference on computer aided control system
44 design, 2006 IEEE international conference on control applications,
45 2006 IEEE international symposium on intelligent control. IEEE, 2006:
46 1645-1650.
[30] Yoo D, Yau S S T, Gao Z. Optimal fast tracking observer bandwidth of
47 the linear extended state observer[J]. International Journal of Control,
48 2007, 80(1): 102-111.
49 [31] Yong F, Yu X, Man Z. Non-singular terminal sliding mode control of
rigid manipulators[J]. Automatica, 2002, 38(12):2159-2167.
50 [32] Taghirad H D. Parallel robots: mechanics and control[M]. Crc Press,
51 2013.
52 [33] Dong B, Shi Y. Prescribed performance synchronization for time ‐
53 delayed complex dynamical networks under event‐triggered pinning
control[J]. International Journal of Robust and Nonlinear Control, 2021,
54
31(18):8989-9007.
55
56
57
58
59
60

You might also like