Upload
singh-mahendra
View
239
Download
5
Embed Size (px)
Citation preview
Intelligent Controller for Mobile Robot Based on Heuristic rule Base Network
Singh Mukesh Kumar1,a, Mishra Deepak Kumar 2,b, Parhi Dayal R.3,c and Singh Mahendra Prasad4,d
1Department of Mechanical Engineering,Govternment Engineering College ,Bilaspur (C.G.), India.
2Department of Mechanical Engineering, Ravishankar Institute of Technology & Management,Raipur (C.G.), India.
3Department of Mechanical Engineering,National Institute of Technology ,Rourkela (Orissa), India.
4Department of Mechanical Engineering, Priyadrashani College of Engineering, Nagpur (MS.), India.
a e-mail:[email protected] , b e-mail:[email protected], c e-mail:[email protected], d e-mail:[email protected]
Keywords- Mobile robot, rule base network, back propagation algorithm.
Abstract— This paper is related to the human perception based idea by using heuristic information
for the navigation of mobile robots in cluttered dynamic environments which provides a general,
robust, safe and optimized path. The heuristic rule base network consists of a simple algorithm
which makes predefined estimation function very smaller. The estimation function should be
adequately defined for desired movement in the environments. A navigation system using rule
based technique that allows a mobile robot to travel in an environment about, which the robot has
no prior knowledge. This heuristic rule is applied in conjunction with artificial neural network. The
proposed intelligent controller provides an optimum trajectory which increases the effectiveness of
a mobile robot. A series of simulations test has been conducted to show the effectiveness of the
proposed algorithm.
Introduction
The current robot navigation systems require intelligent controller able to solve complex problems
under very uncertain and dynamic environmental situations. Navigation algorithms have been
investigated by many researchers [1-3]. Their methods are generally good for local navigation but
may not be optimal in a global sense. To navigate in an unknown or unstructured environment, it is
more desirable for the mobile robot to take intelligently decision based on its sensory information
[4]. In order to adapt the robot’s behavior to any complex, and dynamic environment without
further human intervention, it should be able to extract information from the environment
heuristically, to perceive, and act within the environment. As a result, designing of intelligent
controller of a mobile robot plays an important role in robotics and has thus attracted the attention
of researchers recently [5-7]. Existing approaches plan an initial path based on known information
about the environment, then modify the plan locally as the robot travels or reschedule the entire
path as the robot discovers obstacles with its sensors, sacrificing optimality or computational
efficiency, respectively [8].
Reactive obstacle avoidance is one of the most desirable characteristics of an autonomous mobile
robot. It is important for the robot to respond promptly to its surroundings, for instance, to avoid
unexpected obstacles and continue traveling toward the target [9]. A fuzzy context-dependent
blending of schemas has been proposed by Huq et al. [10]. Their comprehensive result eliminates
the drawbacks of motor schema approaches and provides collision free goal oriented robot
navigation. The Radio Frequency Identification technology algorithm is capable of reaching a target
point in its a priori unknown workspace without vision system, as well as tracking a desired
trajectory with a high precision [11]. The trajectory tracking and dynamic obstacle avoidance of a
Advanced Materials Research Vols. 403-408 (2012) pp 4777-4785Online available since 2011/Nov/29 at www.scientific.net© (2012) Trans Tech Publications, Switzerlanddoi:10.4028/www.scientific.net/AMR.403-408.4777
All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of TTP,www.ttp.net. (ID: 132.174.255.116, University of Pittsburgh, Pittsburgh, USA-26/11/14,19:33:16)
car-like mobile robot within distributed sensor-network spaces is developed by Hwang et al. [12]
and a sequence of experiments is carried out to confirm the performance of the proposed control
system. The general aim of automation is to avoid human interventions to control and to supervise
tasks. Ideally, it should be possible that robot can communicate with technical systems in a similar
way as they do with human assistants [13]. The speed function is proposed such that the minimum
cost path between the starting and target locations in the environment is the optimum planned path.
The speed function is controlled by one parameter, which takes one of three possible values to
generate the safest, the shortest, or the hybrid planned path [14]. The hybrid path is much safer than
the shortest path, but shorter than the safest one.
This paper introduces a control system for a mobile robot which provides heuristic learning
concretely. The method is simple and fast in execution using the perception based heuristic rule
concept. The algorithm computes the paths for the individual robots in the configuration-time space.
Thereby it trades off the distance to static objects as well as with other robots and the length of the
path to be traveled. Useful heuristic rules are hybridized with the ANN to build the desired mapping
between perception and motion. ANN consisting of four inputs the left obstacles distances
(Left_obs), right obstacles distances (Right_obs), front obstacles distances (Front_obs) and the
interim steering angle and an output of final steering angle. A series of simulation test has been
conducted to verify the effectiveness of the controller. In different simulation environment the
proposed approach is well suited to control the motions of a team of robots in a typical environment
and illustrate its advantages over other techniques developed so far.
PERCEPTION BASED HEURISTIC RULE
The heuristic rules are based on human perception (i.e. the working environment provides a fixed
referential frame for the rules). The sensor devices allow to measures the normal ambient light,
which is strongly influenced by the robot’s environment. Based on these sensors a human
perception based heuristic rule can be formulated. The methodology is a general, robust, and safer
which provides fast path planning framework for robotic navigation. If the target is located right
side then the robot will steer clockwise direction i.e. positive steering angle but it the target is
located left side then the robot will steer counter clock direction i.e negative. Human perception
based some of the heuristic rules based on Fig. 1(a) depicts the environment to avoid the obstacles
and motion control by the robots. Based on this scenario some formulated heuristic rules are listed
in Table 1. These parameters can be used to generate different perception based heuristic rules, for
example rule 1 is given below.
TABLE I. HUMAN PERCEPTION BASED HEURISTIC RULE FORMATION FOR OBSTACLE
AVOIDANCE FIG. 2(A)
R.N Left_obs
(mm)
Right_obs
(mm)
Front_obs
(mm)
Tar_ang
(Degree)
Steering angle
(Degree)
1 600 150 600 7 7
2 700 150 150 0 -80
3 515 150 550 90 50
4 420 150 440 53 53
Rule 1: If left_obs = 600 mm and right_obs = 150 mm and front_obs =600 mm and tar_ang = 7˚
Then Change in steering angle = 7˚
Similarly Fig. 1(b) illustrates the environment if the robot enters into U shaped wall, in such a
situation, the robot keep on heading towards the goal position. But when it moves towards the goal
position, it also comes closer to the obstacles. Any obstacle-avoidance behavior except wall-
following behavior would make the robot divert from its goal position. To avoid such situation
some of the rules are listed in Table 2. These parameters can be used to formulate heuristic rule for
wall following, for example rule 5 is given below.
4778 MEMS, NANO and Smart Systems
TABLE II. HUMAN PERCEPTION BASED HEURISTIC RULE FORMATION FOR WALL
FOLLOWING FIG. 2(B)
R.N. Left_obs
(mm)
Right_obs
(mm)
Front_ obs
(mm)
Tar_ang
(Degree)
Steering angle
(Degree)
5 680 1160 1220 13 13
6 780 835 200 0 77
7 80 1510 285 -112 90
8 155 1490 660 168 -90
9 250 530 1085 -117 90
10 235 515 No obstacle -54 -54
Rule 5: If left_obs = 680 mm and right_obs = 1160 mm and front_obs = 600 mm and tar_ang =
13˚ Then Change in steering angle = 13˚
(a) (b)
Figure 1. Human perception based rule formation (a).Rule formation for obstacle avoidance
(b).Rule formation for wall following.
The rules used for navigation of mobile robots are generated by induction from examples.
Approximately one thousand five hundred rules are fed into the induction program, within the
Clementine data ROBNAV software package. The examples present the situations encountered by a
robot while moving in a multi-robot, highly cluttered environment, and the actions that each robot
has to take to avoid colliding with other robots as well as with static obstacles.
BACK PROPAGATION ALGORITHMS
The network is trained to navigate by presenting it with 200 patterns representing typical scenarios,
some of which are depicted in Fig. 2. For example, Fig. 3(a) shows a robot advancing towards an
obstacle, another obstacle being on its right hand side. There are no obstacles to the left of the robot
and no target within sight. The neural network is trained to output a command to the robot to steer
towards its left. During training and during normal operation; the input patterns fed to the neural
network comprise the following components:
= Y }1{1 Left obstacle distance from the robot (1a)
= Y }1{2 Front obstacle distance from the robot (1b)
= Y }1{3 Right obstacle distance from the robot (1c)
= Y }1{4 Target bearing of the robot (1d)
Advanced Materials Research Vols. 403-408 4779
These input values are distributed to the hidden neurons which generate outputs given by
) f(V Y {lay}j
{lay}2 = (2)
where }1{lay-
i. {lay}
jii
{lay}
j YWV ∑= (3)
lay = layer number (2 or 3)
j = label for jth
neuron in hidden layer ‘lay’
i = label for ith
neuron in hidden layer ‘lay-1’
{lay}jiW = weight of the connection from neuron i in layer ‘lay-1’ to neuron j in layer ‘lay’
f(.) = activation function, chosen in this work as the hyperbolic tangent function
xx
xx
ee
ee f(x)
−
−
+
−= (4)
During training, the network output actual may differ from the desired output θdesired as specified
in the training pattern presented to the network. A measure of the performance of the network is the
instantaneous sum-squared difference between θdesired and θactual for the set of presented training
patterns
2actualdesired
sng patternall traini
rr )θ(θ2
1 E −= ∑ (5)
Left_obs
Interim
Steering
Angle
Final
Steering
Angle
Front_obs
Right_obs
Rule Layer
IF…THEN…
Left_obs
Right_obs
Front_obs
Tar_ang
Input
Layer
Output
First hidden layer
Second
hidden
layer
Figure 2. Four-layer heuristic rule neural network for robot navigation
4780 MEMS, NANO and Smart Systems
Robot
Target
Robot
Robot Robot Robot
Robot
Robot
Robot Robot Robot
Robot Robot Robot
Target
Target Target
(a)
(j)
(i) (h) (g)
(f) (e) (d)
(c) (b)
(l) (k)
(o) (n) (m
)
Robot
Robot
Figure 3. Some example of training patterns
The error back propagation method is employed to train the network [15]. This method requires
the computation of local error gradients in order to determine appropriate weight corrections to
reduce Err. For the output layer, the error gradient "{4} is
)θ) (θ (V f δ actualdesired}4{
1'}4{ −= (6)
The local gradient for neurons in hidden layer {lay} is given by
( )}1{laykj
}1{lay
kk
{lay}j
'{lay}j Wδ ) (V f δ ++
∑= (7)
The synaptic weights are updated according to the following expressions
1)(t∆W (t) W 1)(tW ijijij ++=+ (8)
and 1}{lay
i{lay}jijij Y∆ηδ (t)α∆W 1)(t∆W −+=+ (9)
where α = momentum coefficient (chosen empirically as 0.2 in this work)
η = learning rate (chosen empirically as 0.35 in this work) t = iteration number, each iteration consisting of the presentation of a training pattern and
correction of the weights. The final output from the neural network is
)f(V θ 41actual = (10)
where }3{
i}4{
i1i
41 YW V ∑= (11)
It should be noted learning can take place continuously even during normal target seeking
behaviour. This enables the controller to adopt the changes in the robot’s path while moving
towards target.
Advanced Materials Research Vols. 403-408 4781
Simulation results and discussions
The series of simulations results have been obtained by using ROBNAV software has been
developed using C++ [16]. To demonstrate the effectiveness and the robustness of the proposed
method, simulation results on mobile robot navigation in various environments are exhibited.
The first priority of the robot is to avoid static as well as dynamic obstacle. When arrays of
sensor receive the information of object (which is too close to the robot), it avoids a collision by
moving away from it in the opposite direction. Collision avoidance has the highest priority for
mobile robot. And therefore, it can override other behaviors; in this case, the obstacle avoidance
behavior is activated when the readings from any sensors are less than the minimum threshold
values as shown in Fig. 4(a). Another special condition appears as the mobile robot detects an
obstacle in the front while the target tracking control mode is on operation. In this case, the fixed
wall following behavior should be performed first, that is, the mobile robot must rotate clockwise or
counterclockwise such that it can align and move along the wall shown in Fig. 4(b). When the
acquired information from the sensors shows that there are no obstacles around robot, its main
reactive behavior is target steer. Rule based intelligent controller mainly adjusts robots motion
direction and quickly moves it towards the target if there are no obstacles around the robot as shown
in Fig. 5(a).
The results from the proposed method for real time navigation of mobile robot have been
compared with the result from fuzzy controller by Pradhan et al. [7]shown in Fig. 5(a). During
comparisons it has been found that the developed rule based controller has performed better than
the fuzzy controller in terms of path and time optimization as shown in Fig. 5(b).
Obstacles
Figure 4. Simulation result of perception based heuristic rule network (a). Simulation result of
obstacle avoidance (b). Simulation result of wall following.
(b) (a)
Figure 5. Simulation results path comparisons of mobile robot using different controller (a)
Fuzzy controller by Pradhan et al. [7] and (b) Proposed heuristic rule network base controller
4782 MEMS, NANO and Smart Systems
(a)
(b)
Figure 6. Simulation results of Ayari et al. [17] collision free goal reaching in learned
environment (b). Simulation results of proposed method collision free goal reaching.
Table 3 shows a comparison between fuzzy, and proposed rule based network approach. From
table 3 the path traveled by robot using fuzzy controller by Pradhan et al. [7] is 13840 mm and time
taken is 14.67 sec. to reach the target, and whereas the proposed intelligent controller based on
heuristic rule network provides total travel path length 5623 mm and time taken is 5.96 sec. only. A
comparison has also been done between the results obtained by Ayari et al. [17] for navigation of
mobile robot in collision free goal reaching in learned environment and the results from the
developed heuristic rule based intelligent controller (Fig. 6). The effectiveness of the developed
controller has been tested in the series of simulation test and they are in good agreement.
TABLE III. RESULTS COMPARISONS WITH DIFFERENT CONTROLLER.
SN Navigation
observations
Comparisons with various approaches
Fuzzy controller by
Pradhan et al.[7] (Fig.
5(a))
Proposed perception based
heuristic rule network
controller (Fig. 5(a))
1 Length of path (in mm) 13840 5623
2 Time taken (seconds) 14.67 5.96
3 Percentage of result
deviation with proposed
approach (%)
46.13% ---
Advanced Materials Research Vols. 403-408 4783
Conclusions
From the theoretical and simulations analyses the methodology is a general, robust, and safest
which provides fast path planning framework for robotic navigation using human perception based
heuristic rule base network controller. The method is simple but efficient tool for mobile robot
navigation, especially in a real world dynamic environment. The proposed methodology is
successfully applied for navigation in dynamic as well as static environments. The robot rapidly
recognizes their surroundings which provide sufficient information for path optimization during
navigation. The training patterns of ANN can be generated by simulation rather than by
experiments, saving considerable time and effort. Comparison of results between the current
developed and with other techniques have shown good agreement. This validates the authenticity of
the proposed technique. This paper depicts a methodology to achieve the co-operation navigation in
pragmatic way.
References
[1] G. C. Luh, Potential Field Based Immune Network for Dynamic Motion Planning of Mobile
Robots.In proceedings of the 1st International IEEE Forum on Strategic Technology, pp. 151-
155, 18-20 Jan. 2006.
[2] D.R. Parhi, S.K. Pradhan, A.K. Panda, and R.K.Behra, The stable and precise motion control
for multiple mobile robots. Appl. Soft Comput., vol.9 no.2, pp. 477-487, 2009.
[3] D.R. Parhi and M.K. Singh, Intelligent fuzzy interface technique for controller of mobile robot.
J. of Mech. Engineering part C, vol.222 no.11, pp.2281-2292, 2008.
[4] K.T. Song and C.C. Chen, Application of Heuristic Asymmetric Mapping for Mobile Robot
Navigation Using Ultrasonic Sensors. J. of lntelli. and Robot. Sys., vol.17, pp. 243-264, 1996.
[5] D.R. Parhi and M.K. Singh, Real time navigational control of mobile robots using an artificial
neural network. J. of Mech. Engineering part C, vol.223 no.7, pp.1713-1725, 2009.
[6] M.K. Singh and D.R. Parhi, Intelligent neuro-controller for navigation of mobile robot. In the
International Conference on Comput. Intell. (ICAC3-09), Mumbai, India, pp. 123-128, 23-24
January 2009.
[7] S.K. Pradhan, D.R. Parhi and A.K. Panda, Fuzzy logic techniques for navigation of several
mobile robots. Appl. Soft Comput., vol. 9, pp.290-304, 2009.
[8] A.Sipahioglu, O. Yazici, Parlaktuna and U. Gurel, Real-time tour construction for a mobile
robot in a dynamic environment. Robot. Auton. Syst., vol.56, pp.289–295,2008.
[9] K.T. Song, and L.H. Sheen, Heuristic fuzzy-neuro network and its application to reactive
navigation of a mobile robot. Fuzzy Sets and Syst., vol.110, pp.331-340, 2000.
[10] R. Huq, G. K.I. Mann and R. G. Gosine, Mobile robot navigation using motor schema and
fuzzy context dependent behavior modulation. Appl. Soft Comput., vol. 8, pp.422–436, 2008.
[11] W. Gueaieb, and M.S. Miah, An Intelligent Mobile Robot Navigation Technique Using RFID
Technology. IEEE Trans. Instr. and Measure.,vol. 57 no.9, 1908-1917, 2008.
[12] C.L. Hwang and N.W. Chang, Fuzzy Decentralized Sliding-Mode Control of a Car-Like
Mobile Robot in Distributed Sensor-Network Spaces. IEEE Trans. Fuzzy Syst., vol.16 no.1, pp.
97-109, 2008.
[13] A. Kurz , ALEF: An autonomous vehicle which learns basic skills and constructs maps for
navigation. Robot. Auton. Syst., vol.14, pp.171-183, 1995,.
4784 MEMS, NANO and Smart Systems
[14] M. S. Hassouna, A. E. A. Hakim and A. A. Farag, PDE-based robust robotic navigation. Image
and Vision Computing, vol. 27, pp.10–18, 2009.
[15] S. Haykin, Neural Networks a Comprehensive Foundation, 2nd edition, 1999 (Prentice Hall,
Englehood Cliffs, NJ).
[16] D.R. Parhi, Navigation of multiple mobile robots in an unknown environment. Doctoral Thesis,
Cardiff School of Engineering, University of Wales, UK, 2000.
[17] I. Ayari and A. Chatti, Reactive control using behavior modeling of a mobile robot. Int. J. of
Computers, Communications & Control, vol. 2 no.3, 217-228, 2007.
Advanced Materials Research Vols. 403-408 4785
MEMS, NANO and Smart Systems 10.4028/www.scientific.net/AMR.403-408 Intelligent Controller for Mobile Robot Based on Heuristic Rule Base Network 10.4028/www.scientific.net/AMR.403-408.4777