Você está na página 1de 6

2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET)

Model predictive control of a two-link robot arm

El-Hadi GUECHI Samir BOUZOUALEGH Lotfi MESSIKH Sašo BLAŽIC


Automatic Laboratory of Skikda Automatic Laboratory of Skikda Automatic Laboratory of Skikda Faculty of Electrical
26 route El-hadeik, Skikda, 26 route El-hadeik, Skikda, 26 route El-hadeik, Skikda, Engineering
Algeria Algeria Algeria Trzaska 25, Ljubljana, Slovenia
e.guechi@univ-skikda.dz samir.bouzoualegh@gmail.com l.messikh@univ-skikda.dz saso.blazic@fe.uni-lj.si

Abstract—This paper presents a new approach for a model for simultaneous motion planning and control. In [9], kinematic
predictive control dynamics of a two-link manipulator robot. modeling and control of a robot arm using unit dual
This technique consists of linearizing a nonlinear dynamic model quaternions is proposed. In [10], a new approach to tracking
of the robot by using a feedback linearization control. Once, the control of a six degrees of freedom (6-DOF) robotic arm is
linear model has been obtained, a predictive control approach developed. In [11], a convex optimization approach is
will be developed. We have introduced a quadratic criterion and developed for time-optimal path-constrained trajectory
these parameters are calculated to have a specific behavior of the planning of robot systems. In [12], a minimum time control of
closed loop system. Here, the objective is to control the arm robot the Acrobot is proposed. The principle of this approach is to
from an initial configuration to the final configuration using a
use a direct search algorithm for finding an optimal trajectory
predictive control approach and it is obtained by minimizing a
quadratic criterion. In order to show the efficiency of the
for the Acrobot.
proposed method, some simulation results are given. This work deals with a model predictive control (MPC) of a
two link robot arm. Many MPC-based approaches are available
Keywords—Two DOF robot arm; Dynamic model; Nonlinear in the literature. In [13], three different nonlinear model
control; Model predictive control. predictive control algorithms are tested using a fabricated
planar 2-link vertical robotic arm: adaptive nonlinear model
I. INTRODUCTION predictive control (nMPC), PID based nMPC (PIDnMPC), and
Due to the development of production and to facilitate the a novel simplified nMPC (SnMPC). In [14], a novel control of
difficult and repetitive tasks for human has motivated the a manipulator mounted on an unmanned satellite is developed.
increasing interest in arm robotics [1-2]. The mechanical This control consists of two modules: trajectory planning
structure of this class of robots is complex (articulated rigid- module (based on trajectory optimization algorithm) and model
body) which makes the task of control more difficult. Many predictive controller. In [15], a generalized predictive
approaches are available in the literature to control the robotics controller of a flexible single-link manipulator robot is
arms. In [3], a feedback controller is developed to improve the developed and compared with results obtained using a fuzzy
robust performance under structural and parametric uncertainty supervisory controller. In [16], an innovative controller for
disturbance in electro-hydraulic servo system (EHSS) for a 2- flexible-links mechanism based on MPC with constraints is
DOF robotic arm. In [4], a robust control of a robotic arm is proposed. The reference mechanism chosen to evaluate the
developed which takes into account the friction in the robot effectiveness of this control strategy is a four-link closed loop
model. In [5], a coordinated fuzzy control is developed for planar mechanism laying on the horizontal plane driven by a
robotic arms with actuator hysteresis and motion constraints. torque-controlled electric actuator.
Also, the adaptive control scheme is introduced to reduce the This paper proposes a novel approach for controlling a two-
harmful effects from unknown nonlinearities. In [6], a solution link arm robot. This approach involves determining model
to the inverse kinematics problem of a three-link planar predictive control dynamics of a manipulator robot with two
manipulator, needed for generating desired trajectories in the degrees of freedom (DOF). The dynamic model of this robot is
Cartesian space (2D) is found by using a feed-forward neural nonlinear, so a feedback linearization control is applied to the
network. In [7], a hybrid controller for three-degrees-of- robot dynamic model to make it linear. Next, based on the
freedom (3-DOF) robotic manipulators is presented. The obtained linear model, an MPC controller is developed and a
proposed controller comprises of an independent joint quadratic criterion is minimized.
controller, designed in the configuration space, and a sliding
mode controller that enforces desired dynamics for the tracking This paper is organized as follows: in the second section we
error projections to the Frenet-Serret frame. In [8], a novel provide the description of the manipulator robot with two DOF
methodology for motion specification and robust reactive and its dynamic model. In the third section, the control
execution for an industrial robotic manipulators is developed. approach for controlling the robot from an initial configuration
Traditional trajectory generation techniques and optimisation- to the final configuration using model predictive control (MPC)
based control strategies are merged into a unified framework is presented. Simulation results are presented in the fourth
section.

978-1-5386-4449-2/18/$31.00 ©2018 IEEE.

409
2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET)

II. DYNAMIC MODEL where L and τ = [τ 1 τ 2 ] are respectively the Lagrangian of


T

A plane robot with two degrees of freedom can be the motion and the torques vector.
presented as depicted in Fig. 1 Developing the equation (5), the dynamic model of a robotic
arm with two degrees of freedom (DOF) is given by the
following formula [17]:
Y
( )
 M (θ ) θ + C θ , θ + G (θ ) = τ
 (6)
g Y = θ
where :
θ = [θ1 θ 2 ] is joint variable vector;
T
M1 •
θ2
τ = [τ 1 τ 2 ] is torque vector (control input);
T

θ1 • Y is the output vector;
L2 M2 − ( M 1 + M 2 ) gL1 sin (θ1 ) − M 2 gL2 sin (θ1 + θ 2 ) 
y • G (θ ) =  
L1  − M 2 gL2 sin (θ1 + θ 2 ) 
is a vector of gravity torques;
x
X
Fig. 1. Two link robot arm
• ( )
C θ ,θ = 
( )
 − M 2 L1 L2 2θ1θ2 + θ12 sin (θ 2 ) 
 represents
 − M 2 L1 L2θ1θ2 sin (θ 2 ) 
where θi, Li and Mi {i =1,2} are respectively the joint angle, the vector of Coriolis and centrifugal forces;
length and the mass of the first link (i=1) and the second link
(i=2). The gravitational force is noted g.  D D2 
• M (θ ) =  1  is the inertia matrix;
The calculation of the dynamic model of this robot is based  D3 D4 
on the kinetic and potential energies. These are computed using with:
the direct geometric model (DGM) given by the following D1 = ( M 1 + M 2 ) L12 + M 2 L22 + 2 M 2 L1L2 cos (θ 2 )
formula:
D2 = M 2 L22 + M 2 L1L2 cos (θ 2 )
 x = L1 sin (θ1 ) + L2 sin (θ1 + θ 2 ) D3 = D2
 (1)
 y = L1 cos (θ1 ) + L2 cos (θ1 + θ 2 ) D4 = M 2 L22

Using the equation (1), the total kinetic energy of the two link
III. CONTROLLER DESIGN
robot arm is given by the following equation:
1 1 In this section, a predictive control of a robotic arm with two
E = ( M 1 + M 2 ) L12θ12 + M 2 L22θ12 + M 2 L22θ1θ2 + DOF is developed. For that, we consider the nonlinear
2 2
(2) dynamic model given by Eq. (6). First, we determine a
1
2
2 2   ( )
M 2 L2θ 2 + M 2 L1 L2 θ1θ 2 + θ1 cos (θ 2 )
2 feedback linearization control to make the model (6) linear.
Once the linear model has been obtained, a model predictive
and the potential energy is given by the following formula: control will be designed in the second step.

U = M 1 gL1 cos (θ1 ) + A. Feedback Linearization Control


(3)
M 2 g ( L1 cos (θ1 ) + L2 cos (θ1 + θ 2 ) ) The main idea of this technique is to transform the nonlinear
dynamics of the system to one completely or partially linear,
To find the robot motion equations, we use the formalism of such that linear control approaches can be applied to stabilize
Lagrange: it [18-19].
L = E −U (4) Here, the control approach with feedback linearization is
developed for a dynamic model (equation(6)) of the two-link
With the Lagrangian L, we can solve the Euler-Lagrange
robot arm. So, we differentiate the output Y until the control
equation which relies on the partial derivative of kinetic and
input τ appears. In our case, the control input τ appears in the
potential energy properties of mechanical systems to compute
second derivative of the output Y. This implies that the relative
the equations of motion and is defined as follows:
degree is equal to two. The second derivative of Y is given by
d  ∂L  ∂L the following formula:
τ =   − (5)
dt  ∂θ i  ∂θ i

410
2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET)

( ( ) )
Y = θ = M (θ ) −C θ , θ − G (θ ) + τ = v
−1
(7) Now, given the reference angle of the first link θ1d (constant),
the proposed one-horizon time quadratic cost function for
where : stabilizing the system is defined by:
v = [ v1 v2 ] is a synthetic control vector.
T
J = e12 ( t + h ) + ρ e12 ( t + h ) (13)
Finally, from (7) we get the feedback linearization control
where e1 ( t + h ) = θ1d − θ1 ( t + h ) is the predicted angle error,
( )
τ = M (θ ) v + C θ , θ + G (θ ) (8)
e1 ( t + h ) = 0 − θ1 ( t + h ) is the predicted velocity error. The
Applying the control law given by (8) to the nonlinear horizon time h and the weight factor ρ are both positive
system(6), the dynamic model of the manipulator robot with
two DOF, becomes a linear system double integrator. The parameters to be determined later.
relative degree is equal to two. This means that by using the Substituting the prediction model (12) into (13) and
control law (8), we obtain a complete linearization of the minimizing the criterion J with respect to v1 the obtained
nonlinear system (6) and we get a linear system for each joint MPC controller, is given by the following formula:
variable
θ1 ( p ) 1 θ ( p) 1 v1 ( t ) = k3θ1d − k1θ1 ( t ) − k2θ1 ( t ) (14)
= 2 and 2 = (9)
v1 ( p ) p v2 ( p ) p 2
where the following control gains are proposed:
where p is a Laplace variable. 2 2h 2 + 4 ρ
k1 = k3 = 2 and k2 = 3 (15)
The linearization of the nonlinear system has been done. So, h + 4ρ h + 4ρ h
we can develop minimum time control for the two-link robot
arm which will be the goal of the next subsection. The block diagram of the closed-loop system can be presented
as depicted in Fig. 2

B. Model Predictive Control θ1d


v1 1 θ1
In the case of a robot arm with two DOF and after k3
application of the feedback linearization (8) to the nonlinear _ p2
system (6), we obtain the following two decoupled linear
systems. + k1
θ1 = v1
  (10) +
θ 2 = v2
k2 p
Considering the first equation of (10). This system can be
rewritten in the state-space form: Fig. 2. Closed-loop system
 x1 ( t ) = x2 ( t )

 x2 ( t ) = v1 ( t ) (11) Concerning the response of the system, we would like to have
 a behavior similar to a system of the second order:
Y ( t ) = x1 ( t ) w02
(16)
where: p + 2ζ w0 p + w02
2
T
[ x1 x2 ] = θ1 θ1  , v1 is a synthetic control of the first
T

where ζ is a damping factor and w0 is a natural frequency.


link of the robot and Y is the output. The transfer function of the system presented in Fig. 2 is given
Now, we are going to develop a model predictive controller by:
(MPC) (see [20-22]) for the first link of the robot arm. The
MPC controller for the second link of the robot will be
θ1 ( p ) k3
= 2 (17)
developed in the same way as the first. θ1d ( p ) p + k2 p + k1
Assuming v1 ( t ) = v1 constant in the time interval [t t + h ] ,
Taking into account (15), (16) and (17) we obtain:
where h is the horizon time of prediction and using the 2h 2 + 4 ρ
equation (11), we get the prediction model as follows: 2ζ w0 = 3 (18)
h + 4ρ h
θ1 ( t + h ) = v1 h + θ1 ( t )

 1 2 (12) 2
w02 = (19)
θ1 ( t + h ) = v1h + θ1 ( t ) h + θ1 ( t ) h + 4ρ
2
 2
From Eq. (19), the weight factor is:

411
2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET)

2 − ( w0 h )
2 IV. SIMULATION RESULTS
ρ= (20) In order to show the efficiency of the proposed approach,
4w02
some simulation results are given. For simulation purpose, we
Substituting Eq. (20) into Eq. (18), we get the following assume that the mass and the length of the first and the second
second order equation: links of the robot arm are M i =(1,2) = 1( kg ) and Li =(1,2) = 1( m ) ,
w02 h 2 − 4ζ w0 h + 2 = 0 (21) respectively. The initial and the desired orientations of the first
where h is the variable to look for. and the second links of the robot arm are θ1 ( 0 ) = π 2 ,
Choosing ζ = 2 2 and for different values of w0 , we θ 2 ( 0 ) = − π 2 , θ1d = − π 2 and θ 2 d = π 2 respectively.
determine the parameter h from (21). Now, the horizon time is According to the obtained horizon time and the weight factor,
obtained, we use the Eq. (20) to determine the weight factor the gains of the MPC controller are:
with h and ρ being both real and positive. k1 = 15.99; k 2 = 5.65; k3 = k1
The performances of the system are computed for each
determined value of h and ρ and they are summarized in Table
1
2 final position of
the end effector
1.5
TABLE I. SYSTEM PERFORMANCES FOR EACH VALUE OF h AND ρ

position with respect to Y-axis (m)


1 trajectory of the

w0 ( rad s ) 1 4 8 9 end effector

0.5
h ( s) 1.41 0.35 0.17 0.15
initial position of
ρ 1.82×10-8 1.14×10-9 2.85×10-10 1.66×10-10
0 the end effector

min (τ 1 ) -0.5
-26.28 -120.50 -422.01 -529.16 robot base
( Nm ) -1

max (τ 1 )
-1.5
21.15 53.66 170.77 212.24
( Nm )
-2
min (τ 2 ) -2 -1.5 -1 -0.5 0 0.5 1 1.5 2
position with respect to X-axis (m)
-1.76 -28.53 -115.68 -145.80
( Nm )
Fig. 4. Final situation: the robot end effector reaches its objective point.
max (τ 2 )
1.21 19.23 78.32 99.21
( Nm )
trθ1± 5% ( s ) 6.19 1.55 0.77 0.68 2
θ1
trθ±2 5% ( s ) 6.19 1.55 0.77 0.68 1.5 θ1d

Dθ1 ( % ) 13.56 13.58 13.58 13.58 1

Dθ2 ( % ) 13.56 13.58 13.58 13.58


0.5
θ1 and θ1d (rad)

where trθ±i 5% and Dθi {i =1,2} are respectively the settling time
0

within the band (±5%) of the final value and the overshoot in -0.5

(%) of the first joint angle (i=1) and the second joint angle -1
(i=2).
Analyzing the Table1, we observe that the horizon time h has -1.5
an influence on the variation range of the control signals
(robot torques). When h is small then the variation range of -2
0 0.5 1 1.5 2 2.5 3
the robot torques is big. time (s)

For an acceptable control signals range ( max ( τ i ) for Fig. 5. Real and the desired orientation of the first link of the robot
i = {1, 2} ) [17], we choose the horizon time h= 0.35 (s) and
the weight factor ρ=1.14×10-9 which correspond to w0=4
(rad/s). With these chosen parameters, we obtain a better
system performance than in [17].

412
2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET)

2 4
e1
1.5 3 e2
θ2
1 2
θ2d

0.5 1
θ2 and θ2d (rad)

e1 and e2 (rad)
0 0

-0.5 -1

-1 -2

-1.5 -3

-2 -4
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
time (s) time (s)

Fig. 6. Real and the desired orientation of the second link of the robot Fig. 9. Errors in the joint angles

60
v1 Fig. 4 represents the trajectory of the end effector of the
40
v2 robot arm with the final position of both joints depicted. Figs. 5
and 6 represents the convergence of the joint angles θ1 and θ 2 ,
20
respectively, to their reference values. We can notice fast and
asymptotic convergence of both joint variables.
v1 and v2 (rad/s 2)

0 In Fig. 7 the robot synthetic controls v1 and v2 given by the


Eq. (14) are shown. As we can see, the synthetic controls reach
-20 zero when the end effector of the robot reaches its objective.
Fig. 8 shows the robot torques that can be obtained from
synthetic controls using Eq. (8). The convergence of the joint
-40
angles errors e1 and e2 of the two-link of the robot arm
towards zero using the proposed approach of control is
-60
0 0.5 1 1.5 2 2.5 3 depicted in Fig. 9.
time (s)

Fig. 7. Robot synthetic controls V. CONCLUSION


The present article proposes a model predictive
control (MPC) approach for a robotic arm with two degrees of
60 freedom. This technique consists of linearizing a nonlinear
τ1 dynamic model of the robot by using a feedback linearization
40
τ2 control. Next, based on the obtained linear model, an MPC
20 controller has been developed. The time h and the weight
0
factor ρ are computed, so that the closed loop system have a
behavior of a second order system. The obtained simulation
-20 results show the efficiency of the proposed approach. Future
τ1 and τ2 (Nm)

-40 work is aimed towards introducing the control horizon in the


cost function to further improve the system performances.
-60
After that, the validation of the proposed approach on a real
-80 robot is envisaged.
-100

-120 ACKNOWLEDGMENT
-140 This work was supported by the Ministry of Higher Education
0 0.5 1 1.5
time (s)
2 2.5 3
and Scientific Research of Algeria (CNEPRU J0201620140014
). The authors would like to thank the reviewers for their
Fig. 8. Robot torques valuable comments.

413
2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET)

REFERENCES [21] P. J. Gawthrop, L. Wang, “Intermittent predictive control of an inverted


pendulum,” Control Engineering Practice, vol.14, no.11, pp.1347‐1356,
[1] M. Zheng, W. Liao, C. Yin and A. Wang, "Nonlinear Tracking Control 2006.
Design of A Robot Arm Using Robust Right Coprime Factorization and
Sliding Mode Approaches", Proceedings of the 2014 International [22] A. Mills, A. Wills, B. Ninness, “Nonlinear model predictive control of
Conference on Advanced Mechatronic Systems, pp. 11-16, August an inverted pendulum”, Proceedings of the American Control
2014, Japan. Conference, St. Louis, pp. 2335‐2340, 2009.
[2] V. Zanotto, A. Gasparetto, A. Lanzutti, P. Boscariol, and R. Vidoni,
"Experimental Validation of Minimum Time-jerk Algorithms for
Industrial Robots", Journal of Intelligent & Robotic Systems, vol. 64,n°.
2, pp. 197-219, November 2011.
[3] Q. Guoa, T. Yub, and D. Jiangc, "Robust H? positional control of 2-
DOF robotic arm driven by electro-hydraulic servo system", ISA
Transactions, vol. 59, pp. 55-94, November 2015.
[4] M. Plooij, W. Wolfslag, and M. Wisse, "Robust feedforward control of
robotic arms with friction model uncertainty", Robotics and
Autonomous systems, vol. 70, pp. 83-91, August 2015.
[5] Z. Liua, C. Chena, Y. Zhanga, and C.L. Philip Chenb, "Coordinated
fuzzy control of robotic arms with actuator nonlinearities and motion
constraints", Information Sciences, vol. 296, pp. 1-13, March 2015.
[6] A-V. Duka, "Neural Network based Inverse Kinematics Solution for
Trajectory Tracking of a Robotic Arm", Procedia Technology, vol. 12,
pp. 20-27, 2014.
[7] T. Uzunovic, E-A. Baranb, E. Golubovicc, and A. Sabanovicd, "A novel
hybrid contouring control method for 3-DOF robotic manipulators",
Mechatronics, vol. 40, pp. 178-193, December 2016.
[8] A.M. Zanchettin, and P. Rocco, "Motion planning for robotic
manipulators using robust constrained control", Control Engineering
Practice, vol. 59, pp. 127-136, February 2017.
[9] E. Özgüra, and Y. Mezouarb, "Kinematic modeling and control of a
robot arm using unit dual quaternions", Robotics and Autonomous,
Systems, vol. 77, pp. 66-73, March 2016.
[10] E.D. Markusa, J.T. Ageeb, and A.A. Jimohc, "Flat control of industrial
robotic manipulators", Robotics and Autonomous systems, vol. 87, pp.
226-236, January 2017.
[11] D. Verscheure, B. Demeulenaere, J. Swevers, J. De Schutter, and M.
Diehl, "Practical Time-Optimal Trajectory Planning for Robots: a
Convex Optimization Approach", IEEE Transaction on Automatic
control, vol. 54, n°. 10, pp. 2318-2327, 2009.
[12] G. Boone, "Minimum-time Control of the Acrobot", In Proc IEEE
Intemational Conference on Robotics and Automation, New Mexico,
1997, pp. 3281-3287.
[13] J. Wilson, M. Charest and R. Dubay, “Non-linear model predictive
control schemes with application on a 2 link vertical robot manipulator”,
Robotics and Computer-Integrated Manufacturing, vol. 41, pp. 23–30,
October 2016.
[14] T. Rybus, K. Seweryn and J. Z. Sasiadek, “Control system for free-
floating space manipulator based on nonlinear model predictive control
(NMPC)”, Journal of Intelligent & Robotic Systems, vol. 85, n°. 3, pp.
491-509, March 2017.
[15] R. Boucetta, “Generalized Predictive Control for a Flexible Single-Link
Manipulator”, Computer Information Systems and Industrial
Management, pp. 499-510, 2013.
[16] P. Boscariol, A. Gasparetto and V. Zanotto, “Model Predictive Control
of a Flexible Links Mechanism”, Journal of Intelligent and Robotic
Systems, vol. 58, n°. 2, pp. 125-147, May 2010.
[17] I. David, and G. Robles, "PID control dynamics of a Robotics arm
manipulator with two degrees of Freedom", Control de Processos y
Robotica, pp. 3-7, 17th August 2012.
[18] H. K. Khalil, "Nonlinear systems - Third edition", Prentice Hall Inc,
New Jersey, 2002.
[19] P. Farzin, M.H. Yarmahmoudi, M. Mirzaie, S. Emamzadeh, and Z.
Hivand, "Design novel fuzzy robust feedback linearization control with
application to robot manipulator", I.J. Intelligent Systems and
Applications, 05, pp.1-10, 2013.
[20] L. Magni, R. Scattolini, K, Aström, “Global stabilization of the inverted
pendulum using model predictive control,” Proceedings of the 15th
IFAC World Congress, Barcelona, 2002.

414

Você também pode gostar