2017 International Conference on Robotics and Automation Sciences

Mecanum Wheels with Astar Algorithm and Fuzzy PID Algorithm Based on
Genetic Algorithm

Rongyan Wen Minming Tong

Department of Electrical Engineering Department of Electrical Engineering
China University of Mining and Technology China University of Mining and Technology
Xuzhou, 221008 Jiangsu, China Xuzhou, 221008 Jiangsu, China
e-mail: [email protected] e-mail: [email protected]

Abstract—In the industrial fields, Mecanum robots have been

widely used. The Mecanum Wheel can do omnidirectional
movements by electric machinery drive. It’s more flexible than
ordinary robots. It has massive potential in some situation
which has small space. The robots with control system can
complete the function of location and the calculation of
optimal route. The Astar algorithm is most common mothed.
However, Due to the orthogonal turning point, this algorithm
takes a lot of Adjusting time. The Improved algorithm raised
in this paper can reduce the occurrence of orthogonal turning
point. It can generate a new smooth path automatically. This Figure 1. Mecanum wheel chassis
method can greatly reduce the time of the motion of the path.
At the same time, it is difficult to obtain satisfactory In the kinematic analysis of the wheel, when we calculate
performance by using the traditional control algorithm the relative position of the roller on the wheel relative to the
because of the complicated road conditions and the difficulty of center of the robot, we can calculate the motion of the whole
establishing the model of the robot, so we use fuzzy algorithm robot [2].
to control robots. In fuzzy algorithm, the use of static By Lagrange the second equation:
membership function will affect the control effect, therefore,
for complex control environment, using PSO algorithm to G § w/ · w/ w)
dynamically determine the membership function. It can ¨¨ ¸¸   4L (i=1,2,……..,)
effectively improve the motion performance and improve the
GW © wTL ¹ wT L wTL
dynamic characteristics and the adjustment time of the robot. where L is the Lagrange function, L=V-U.
The mobile robot has three degrees of freedom. We can
Keywords-Mecanum wheels; improved Astar algorithm;
get the omni-directional mobile robot's Lagrange function:
improved fuzzy algorithm based on genetic algorithm
1 1 1
L V U m vx2  v y2  J Z 2  J 2 T1  T2  T3  T 4  U
I. INTRODUCTION 2 2 2
M is the mass, J is the rotational inertia of the robot around
The human’s work environment is getting worse and
the vertical axis, J2 is the moment of inertia of each wheel
worse. In the storage industry, disaster rescue scene and
around its center. The speed of the wheel is:
military field, Mecanum wheel robot is widely used, it has
played a role in detecting and discovering information. But it ­ 5 
can’t achieve satisfactory performance such as demanding °Y [ T  T  T  T
precision, poor operating conditions of the occasion. It is ° 
necessary to improve the robot path planning algorithms and ° 5
control algorithms. So as to reduce the time of motion. ®Y \  T  T  T  T
° 
II. MECANUM WHEEL
° 5
The McNam’s wheel chassis used in this article is a °Z T  T  T  T
combination of two wheels, L-auron and right-handed. Using ¯  O  O
motor drive can achieve omnidirectional mobile [1]. We adjust the movement of the motor by adjusting the
There are two ways as shown in the figure, the left is motor speed. [3]
used in this article. Driving mechanism using the suspension
mechanism can effectively avoid bumps to ensure the III. COMPUTER CONTROL UNIT
smooth operation of the organization, it also makes the latter The processor uses a radio frequency sensor to locate
part of the image processing more simple. itself, but we must embed a few radio frequency cards on the

978-1-5386-3995-5/17/$31.00 ©2017 IEEE 114

ground in advance, they help robots to make decision. The IV. IMPROVED ASTAR ALGORITHM
control system uses a magnetic sensor to provide travel Combining environmental information and their own
information to the robot control system. The computer state information, the system can draw the around map.
connects the magnetic stripe in different locations as needed. Through repeatedly drawing, the robot can acquire more
The computer can also determine the direction of the robot accurate map information. Through the positioning system,
according to the relative position of the robot through the the robot can detect its own location, and then find the target
sensor and the magnetic stripe. Since the omnidirectional point, finally, the system apply a certain algorithm to plan
mobile robot is free to move, two magnetic sensors are used the optimal path. [4] Astar algorithm is a path planning
to determine the state of the robot. As the picture shows: method, it calculates the function values of different paths to
determine the optimal path. It will calculate all nodes of each
path, determining the minimum path. The value of the
function usually uses Euclidean distance and weight as a
parameter.
In this algorithm, the system will fall into the trap of the
deep search situation, which will lead to the slow system
response. We can set the search depth to avoid the
occurrence of this situation.
Another drawback of the algorithm is that this algorithm
ignores the geometric distribution of the real node. In the
optimal path there will be many rectangular node, which will
result in the motor to start and stop. It is not conducive to the
implementation of the system.
There are two ways to solve this problem. Firstly, we can
reduce the weight of the function at rectangular node to
avoid the occurrence of rectangular node in the path.
Secondly, to the path consisting of rectangular node, we can
artificially smooth the path. For example, if we have three
nodes which form right angle, we can apply the improved
Astar algorithm to smooth path. In the end, this can improve
the efficiency of the control system.
Figure 2. Magnetic guidance sensor layout

The controller also uses the visual sensor and the infrared
sensor and the ultrasonic radar to detect and update the map
information.
Ultrasonic sensor gathers information to detect obstacles,
which is aim to avoid accidents.
When the system is running, the information of the
surroundings is firstly detected by the laser sensor, and the
robot starting position is taken as the absolute origin. The
environment information that the robot reads through the
sensor at the origin will be taken as the location of the robot.
When the robot is walking, the robot will calculate the
current relative coordinates through the real-time table Figure 3. A* Algorithm analysis diagram
update and store the environmental information. At the same
time, the robot will re-calibrate the direction of the For node 1:
environment calibration point, until the robot walks again, g1 0
the robot will be in accordance with the newly calibrated h1 Dx * Dx  S x * S x  Dy * Dy  S y * S y
point of their own positioning calculation.
The error caused by sensors will raise fault in a long time f 1 g 1 h1
running situation. Then the system noise will increase. In where f (1) is the function value of the certain node, g (1) is
long time working, the error will increase. The errors will the weight of this node. Then we calculate node 2~9,
result in the uncertainty of the positioning information, we choosing the minimum as the next node. If we determine the
can use the laser calibration method to improve, that is, 4th node as the next node, we will choose node 4 as the basic
repeatedly measure the calibration point to enhance the robot node to determine the following node. The following figure
attitude information to achieve a more accurate map. is an example:

115
 The system uses the fuzzy algorithm to derive the PID
parameters that need to be adjusted. The system adds those
parameters to the drive section. The fuzzy controller uses
two parameters as input, one is the amount of deviation, the
other is the gradient of deviation. Fuzzy rules can be roughly
divided into:
(1) When the deviation is large, the system should
increase the proportion of parameters, and reduce the
differential parameters. The integral parameter should be
adjust to 0.
(2) When both the deviation and the gradient of deviation
change moderately, the system should reduce the integral
parameters.
(3) When the deviation is small, the system should
Figure 4. A* Path planning diagram increase the proportion of parameter. Meanwhile both the
differential parameters and integral parameters should not be
There are a few rectangular nodes. By reducing the changed.
weight, the rectangular node in the path can be reduced. The
robot will design a new smooth path according to those
rectangular nodes. In this way, rectangular nodes are
replaced by a smooth curve, therefore, the robot’s motor no
longer to start and stop repeatedly and the time of control
will reduce to the minimum. Meanwhile, because of the
discard of rectangular nodes, the non-continuity problem
turns into continuity problem. This method can greatly cut
down complexity of the control system and reduce the loss of
the control system’s components.
By using this improved algorithm, according to the result
of experiment, it can save 40% to 60% of the previous time
which contains many rectangular nodes. Hence, this
improved A* method can greatly enhance efficiency of the
robots.
V. IMPROVED FUZZY PID CONTROL
After attaining the improved smooth motion path, the
Figure 6. Fuzzy control rule table
robot needs a reliable and effective control system to ensure
that the robot can accurately track the path. Traditionally,
However, in the case of large disturbance the algorithm is
PID control algorithms are used to control the motor. [5]
not good. One of the important reasons is the membership
However, this mature algorithm requires an accurate
function parameters. If we use static membership function, to
mathematical model. But it is difficult to obtain a precise
the situation of changing of the environment, the effect of
mathematical model due to the complex environment and the
control will not meet our request. What we should do is to
complexity of the robot, so we will use fuzzy PID control
dynamically acquire the membership function according to
algorithm. In the closed-loop feedback structure we add
the actual condition.
fuzzy controller. According to experience, we adjust the
By introducing genetic algorithm into fuzzy PID
parameters of the PID artificially so as to respond quickly to
controller, it can effectively improve the adaptive problem of
the environment. The block diagram of the system design is
membership function. Using the selection, crossover and
as follows:
mutation in the genetic algorithm, the optimal solution in the
Member- solution space can be found, which can effectively improve
Fuzzy-rule Algorithm
ship the system response when the disturbance is large. In other
base base
function words, the fuzzy controller can also combine with the genetic
algorithm to obtain a wide range of fuzzy control rules
strategy.
In this paper, genetic algorithms are used only in
Sensor Fuzzy Anti-
determining membership functions.
system reasoning fuzzification
target Firstly, we must find the encoding of the fuzzy set.
According to the actual situation, we convert the decimal
number into binary numbers, and this binary number as a
Figure 5. Fuzzy control model diagram membership function of the unique identity, as the

116
distinction between other membership function information.
The permutation formula is as follows: f( ) = ( ( ( )− ( , )
( )) )
−
x = (2 − 1 )
− In the formula, there are N output fuzzy subsets, ( )
In the formula, m is the number of bits in the binary is expected output, ( , ) ( ) is actual output, is input.
number,[a, b] is domain. This permutation ignores the error: The third step is genetic operation. In genetic operations,
excellent coding individuals have greater chance of breeding,
− according to the fitness function we can find the fitness
e=
2 −1 function value of the individual, and give the individual
greater genetic probability. Common methods have a fitness
To simplify the discussion process, we assume that the ratio method:
shape of the membership function is a triangle. At the same
time, in order to ensure that any input have a corresponding = ∑
output, we set the membership function to the triangular
membership function. Meanwhile we make sure the function
is probability of selection. In the design of the crossover
line pass through μx=0.5 this point. We should meet the
operator for the reconstruction of the individual structure, the
conditions in fuzzy partion:
crossover operator should meet the crossover operator's
evaluation criterion. The operators need to ensure that the
∀x ∈ X: ( )=1 traits of the outstanding individuals in the previous
generation are inherited as much as possible in the new
individuals of the later generation, and the coding design and
is the membership function of . For a given domain, it the cross design are closely related. The crossover operation
is possible to determine the specific shape of the membership can be divided into the following steps:
function by simply specifying the vertices of the membership (1) We will decode two parent individuals, that is, to
function of the triangle. So we can uniquely determine this restore them to absolute coordinates of the binary
membership function by simply encoding each vertex. code.
x0 x1 x2 x3 (2) Randomly select the intersection position in the
1 locus.
(3) Exchange the value of the locus.
0.5
(4) The values on the L loci were crossed in the
appropriate form.
-10 -8 -6 -4 -2 0 2 4 6 8 10 (5) Re-encode the generated new individual.
Variation is to change the gene values of certain loci into
an individual. In the genetic algorithm, the crossover
Figure 7. Membership function encoding operator is the main operator because of its global search
ability. The mutation operator is used as the auxiliary
= ,x ∈ ( , ) operator because of its local search ability. The genetic
… algorithm is able to balance the global search and local
= ,x ∈ ( , ). search by crossing and mutating the pairing and competing
operation. When the group is in the evolution of the local
space in the search space and only by the cross can’t get rid
In this paper we use the distance between two points to of, through the mutation operation of the above
encode this membership function, such as distance − . characteristics can help this out.
By encoding the distance above, we can get the unique According to the result of experiments, it can be
information of different membership function. We use the concluded that the improved algorithm has good dynamic
convert formula above to convert the decimal value x into characteristics, and the adjustment time is short, the system is
binary numbers. This can facilitate the subsequent genetic more stable.
operation. Assuming that the fuzzy evolution system has m
fuzzy subset language values, only the m-1 coding value can VI. CONCLUSION
be used to calibrate the membership function. At the same In this paper, we use genetic algorithm to adjust the
time, due to symmetry, this method can greatly reduce the membership function of fuzzy subset. We optimize the fuzzy
parameters and improve the efficiency of the system. control and make the control system have the dynamic
Similarly, for the trapezoidal membership function, we can ability. We propose an improved coding method and
also use the encoding method to encode, just increase the crossover operator for the characteristics of membership
number of parameters. [6] function, Speed up the convergence rate. The results show
Secondly, we need to design the fitness function. In that the genetic algorithm is suited to optimize the
general, the fitness function is: membership function, which can make the system have

117
excellent control performance. In order to obtain a more [2] Mohd Salih J E, Rizon M, Yaacob S, et al. Designing omni-
effective motion path, we improve A * path planning directional mobile robot with mecanum wheel.[J]. American Journal
of Applied Sciences, 2006, 3(5).
algorithm to achieve a smooth motion for the robot.
[3] Indiveri G, Swedish Wheeled Omnidirectional Mobile R obots:
Kinematics Analysisand Control[J].IEEE Transactions on Robotics,
ACKNOWLEDGMENT 2009,25(1):164-171.
This work was supported by National key research and [4] Hart P E, Nilsson N J, Raphael B. A formal basis for the heuristic
development program (No. 2016YFC0801808). determination of minimum cost paths[J]. IEEE Transaction on
Systems Science and Cybernetics,1968,4(2):100-107
REFERENCES [5] Roland Siegwart, Illah Reza Nourbakhsh. Introduction to
Autonomous Mobile Robots. MIT Press;2nd Revised
[1] Gferrer A. Geometry and kinematics of the Mecanum
edition(2011.2.18)
wheel[J]Computer Aided Geometric Design,2008,25:785-791J. Clerk
Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol. 2. [6] Goldberg D E .Genetic algorithms in search , optimization and
Oxford: Clarendon, 1892, pp.68–73. machine learning .Addison-Wesley Pub Co , Inc,1989

118