| Login | Become a Member | Add a Quote | Saturday, November 21, 2009 |
| ||
|
| ||
| My Profile | Browse |Search | Submit | Contact |
| Words of Wisdom: | ""Evil is not bad, just missunderstood" - Tom Felton" - Holly |
|
| ||
A New Mobile Robot Navigation Algorithm A New Mobile Robot Navigation Algorithm By using Fuzzy Logic Control and Virtual Target Technique Dr. Omid Reza Esmaeili Motlagh Intelligent Systems and Robotics (ISRL) Lab Institute of Advanced Technology, University Putra Malaysia (UPM), Malaysia Abstract – A new fuzzy logic control system is developed for reactive navigation of a behavior-based mobile robot. The robot perceives its environment through an array of eight sonar sensors and self positioning-localization sensors. While the fuzzy logic body of the algorithm performs the main tasks of obstacle avoidance, target seeking and speed control, an actual-virtual target switch strategy enables the robot to wall following behavior when needed. This significantly results in resolving the problem of limit cycles in any type of multiple dead end in local navigation which is an advantage beyond pure fuzzy logic approach and common virtual target techniques. In this work, multiple traps may have any shape or arrangement from barriers forming simple corners and U-shape dead ends to loops, maze, snail shape, and other complicated shapes. Under control of the algorithm, the robot makes logical trajectories, avoids any obstacle, and adjusts its speed efficiently for better obstacle avoidance and according to power considerations and actual limits. Final trajectory results are demonstrated by simulation work in compression with results from other related methods to show the effectiveness of the proposed approach. 1. Introduction Autonomous mobile robots defined as robots capable of intelligent motion and action without requiring either a guide to follow or a tele-operator control, which involves the integration of many different bodies of knowledge. This makes mobile robotics a challenge worthwhile. Mobile robot local path planning or also called reactive navigation in an unknown and changing environment with uncertainties is one of the most challenging problems in robotics. For real-time autonomous navigation, the robot should be capable of sensing its environment, interpreting the sensed information to obtain the knowledge of its position and the environment, planning a real-time route from an initial position to a target with obstacle avoidance, and controlling the robot direction and velocity to reach the target [1]. There are many approaches to real-time local navigation that deal with one or more related issues and try to tackle the problem in their own way. Some of the online approaches to path planning include wall following method [2], artificial potential field methods[3,4], virtual target approach [5,6], land mark learning [7], edge detection, graph-based methods, vector field histogram methods, dynamic window approaches, neural network based methods, fuzzy logic methods [8,9], and many others. Since the introduction of the fuzzy logic control (FLC) in 1965 by Lotfi Zadeh of the University of California at Berkeley [10], this approach to mobile robot navigation and obstacle avoidance has been investigated by several researchers. Fuzzy systems have the ability to treat uncertain and imprecise information using linguistic rules, thus, they offer possible implementation of human knowledge and experience and have an advantage in that they do not require a precise analytical model of the environment. 1.1. The problem of multiple dead end Although fuzzy logic control (FLC) technique can simplify navigation problems but there are situations happen in local work space where a pure fuzzy logic approach fails in taking appropriate action. Of these troublesome situations obstacles forming a loop shape also called dead end traps are the most common. The local minima situation occurs when a robot navigating past obstacles towards a target with no prior knowledge of the environment gets trapped in a loop. This happens if the environment consists of concave obstacles, and the like. Fig.1 shows a robot with pure fuzzy logic navigator getting trapped in a U-shape dead end. Because rules that are fired for target attractor and obstacle repulsor modules give output actions that neutralize each other, the robot gets into an infinite loop or local minima [7]. Fig.1. Inability of fuzzy logic control in Local Minimum Initially the robot moves directly toward the target due to target seeking behavior, because there is no obstacle sensed in front of the robot and this is an ideal shortest path to the target. This continues until point “A” Where the robot detects obstacles at the direct front, it makes a right turn due to obstacle avoidance behavior that results to wall following until robot reaches the point “C”. This is because until this point both the target and obstacle are at the left hand side of the robot. But as the robot in passing by the point “C” the target is going to be at right hand side of the robot, while the obstacle is still at the left hand side. Therefore at the point “D” due to both target seeking and obstacle avoidance behaviors the robot goes back to ward the target. The result of this behavior is that the robot wanders indefinitely in the dead end trap called the limit cycle problem or local minimum. The situation even worsens when mobile robot encounters two or more dead ends in row forming a multiple minimum. In the next part few related works on minimum avoidance will be review and their draw backs in multiple minimum situations will be discussed. Later in part 3 a new target switch approach to resolve the stated problem will be introduced. 2. Related work Other than pure fuzzy logic approaches there are many types of combinational navigation approaches, which range broadly according to two aspects of the extent to which their fuzzy logic navigator makes ultimate navigation decisions, and their degrees of dependence on memory data and artificial intelligence. 2.1. Zhu and Yang state memory method [8] In this method the “dead cycle” problem is resolved by using a state memory strategy. The variables from which this method makes ultimate decision of either to keep turning around the obstacle (states 1 or 2) or to reach the target (state 0) are the robot current distance to the target (Dc) and the robot initial distance to the target (Dm) memorized when the obstacle was first detected (first change from state 0 to state 1 or 2). But this distance-based decision making results in poor trajectories in situations like here from point “D” toward the target. Since Dm was memorized when the robot was traveling from “A” to “B”, and the condition for changing the state to 0 or target seeking is; Dc shorter than Dm, therefore from point “D” the robot can not go straight toward the target and has to keep turning around the obstacle until point “E” where Dc becomes shorter than Dm. Other than this, large size of the whole algorithm including a set of 48 rule bases and the assistant state memorizing algorithm make the method less interesting. Fig.2. Yang method using distance-based memory states 2.2. Krishna and kalra landmark learning [7] This method is a real-time collision avoidance algorithm with the local minima problem resolved by classifying the environment based on the spatio-temporal sensory data sequences. Although this method has a good result as shown in Fig.3.a but it highly depends on the landmark recognition and therefore needs exact coordination localization. In addition, it is difficult to choose a correct direction to follow the wall boundary as seen in Fig.3.b. [9]. Fig 3. Krishna and kalra landmark learning approach 2.3. Wang and LIU minimum risk method [9] This method is based on “trial-and-return” behavior phenomenon as shown in Fig.4.a and 4.b. “If there is no obstacle blocking the nearest exit, the minimum risk approach is guaranteed to find a nearest exit to escape from the local minimum and ultimately reach the goal”. Therefore when the nearest exit in Fig.4.a is blocked by a long wall based on this approach the nearest exit will be the opening at the right hand side where the wall ends. And based on the same principle the robot can find its way out of the U-shape dead end in Fig.4.b. Fig.4. Wang minimum risk approach based on FL Although the minimum risk approach seems to be robust in traps of this type but obviously there is a major problem with trial and return-based motion because for a mobile robot Power consumption And spent time from start to target are the two important issues that are totally ignored in this approach. This is shown in Fig.5 with an example of a mobile robot trying hardly to find its way out of a simple dead end. Fig.5. trial and error navigation causes loss of time and power. 2.4. Xu and Tso virtual target method [5] Has good properties in minimum avoidance behavior and the robot makes logical trajectories. (Fig.6) Fig.6. Xu. and Tso target switch scheme; virtually shifting the target radian to the opposite direction In this method the target is virtually relocated to points directed away from the true target, therefore after passing through the critical points “C” or “F”, the robot maintains the same tendency of turning to the left from C I J N or to the right from F L M N. The virtual target orientation is given by tr = -(- tr0) following the point C or tr = - tr0 following the point F, where tr0 and tr represent the real and virtual target orientations respectively, and the condition for the target switching is given by |tr1 – tr0| > TR , where tr0 and tr1 are the real target orientations at the two consecutive reaction instants and TR is a threshold for the abrupt change in the target orientation. Due to the actual to virtual target switch applied in this method, robot is able to quit limit cycles and find a way out of the dead end. Even in loop shapes like in Fig.7.a, still robot can find the opening at point “C” due to the fuzzy logic target seeking and obstacle avoidance behaviors. The algorithm shows its major weakness when it fails to reach the goal in such a concave and recursive U-shape environment shown in Fig.7.b. “This is because the robot encounters another local minimum at the location “B” and “C” when it is working under the influence of previous virtual sub goal” [9]. Therefore this method is not applicable in multiple minimum situations. Fig.7. (a) The common target switch algorithm is applicable in simple minimum situation, (b) inability of the algorithm in multiple minimum situations. 2.5. Conclusion From all the methods discussed here the virtual target technique has better properties in that unlike the landmark learning approach, it doesn’t require bulky process of memory integration. Also there is no need to change the basic fuzzy logic rule bases in order to switch from actual to virtual target and vice versus. Using this technique a reactive robot makes more logical trajectories compared to the methods like the minimum risk that take “trial and return” approach. Therefore the idea of actual-virtual target switch will be the main concept in development of a new approach to the minimum problem while a typical fuzzy logic controller will carry out the basic tasks of target seeking and obstacle avoidance. In this project, the main contribution of the work is in solving the problem of multiple minimum traps in local navigation whereas multiple minimum traps are considered as different possible arrangements of bars, barriers or walls forming any dead end shape from U-shape dead ends to loops, maze, snail shapes and even more complicated shapes. Therefore the objective of this project will be to develop a new fuzzy logic based algorithm by using a virtual target technique for robot local navigation in multiple minimum situations. 3. The proposed target switch approach In the new approach the critical point to switch from real target to virtual target is the point where the robot detects the obstacle for the first time. Also there is a difference in the actual to virtual target switch process and vice versus. 4. Experimental results A pioneer 3TM mobile robot has been used to show the performance of the proposed algorithm. In Simulation investigations, the starting location of the robot and the location of the target are given for each navigation task. The normal speed of the robot is 5 cm/s, the upper bound of the robot speed is 10 cm/s. The pioneer 3TM is equipped with an array of 8 sonar range finders ( 2 at sides and 6 at front with 20 degree interval ) to perceive its environment. The Simulation and programming soft ware for Active media robotics graphic Simulator is programmable by using Colbert language which is a limited form of C++ added with Robot self localization functions. Fig.13. The robot behavior in compression with other approaches in the same dead end trap as was shown in figures; 2, 3.a, 5, and 7.b. Fig.13 depicts the expected results as previously discussed in part 3. In this method unlike the memory state method, [8], the robot doesn’t have extra wall following and goes directly toward the target as soon as it is out of the dead end trap. And because the motions is not a trail and return type the robot makes more logical trajectories in compression with the minimum risk approach. Fig.14. Snail shape dead end A more complicated trap of this type is the snail shape shown in figure.14. Here after straight motion from “A” to “B”, the robot encounters the first wall and therefore follows the walls from “B” to “C” in hope of finding an opening. As soon as it detects the opening at point “C” most logical decision is to follow the walls in the opposite direction from “C” to “D” where the robot switches to actual target seeking behavior (case 1). The motion doesn’t have to be in one direction until opening appears and then in the opposite direction. In the example shown in Fig.15, the robot motion is primarily towards the target until the wall is detected and then clockwise counter clockwise clockwise counter clockwise, until it is out of the trap and therefore can reach the real target. More example of this type is shown in fig.16, to prove the robustness of the algorithm in different multiple minimum situations. Fig.15 It was found in experimental results that the robot makes undesirable extra turns if the condition for turning is just based on the sum of the turned angles. In the example of opening snail shape shown in Fig.17, until point “A” the robot has a sum of turning angles equal to 360 degree anti-clockwise. Therefore for algorithm without considering the case 4, the robot would continue to wall following until point “B” where there is no obstacle around and the sum of turned angles is compensated clockwise to zero (Fig.17.a). In order to solve the problem of extra turns a distance based condition is added to the algorithm (case 4). In this case the decision to switch back to the actual target is taken if only there is no obstacle around and the robot distance to the actual target has the minimum amount compared to the distance between the robot and the target at any time during the robot motion. Fig.17.b shows how the robot motion is modified at point “A” to avoid making extra turns. Fig.16. Examples of the robot motion 5. Conclusion A new fuzzy logic control system is developed for reactive navigation of a behavior-based mobile robot. The inputs to the fuzzy controller are; the obstacle position detected relative to the robot heading and the target orientation which is defined as the angle between the robot heading direction and the robot-to-target direction. While the fuzzy logic body of the algorithm performs the main tasks of obstacle avoidance, target seeking and speed control, an actual-virtual target switch strategy enables the robot to wall following behavior when needed. This significantly results in resolving the problem of multiple dead ends in local navigation which is an advantage beyond pure fuzzy logic approach and common virtual target techniques. The work is in progress for the situations where the environment is dynamic and the target shift needs to be supervised by some kind of landmark learning. But still this kind of smooth target shifting seems more interesting because of its similarity with human fuzzy behavior in gradually shifting their direction to left or right in order to turn around an obstacle, and then compensate their turns in opposite direction until finding the original direction to the target. Fig.17. distance based decision making. (a) Without case condition (case 4); undesired urns from “A” to “B”,(b) case condition (case 4); robot-target distance has its minimum amount at point “A”; straight motion to the actual target. 6. References [1]. Kristof Goris, “Autonomous Mobile Robot Mechanical Design “,verije university, brussel, Academiejaar 2004-2005 [2]. Andrew G. Lamperski, Owen Y. Loh, Brett L. Kutscher and Noah J. Cowan Johns Hopkins UniversityBaltimore “ Dynamical Wall Following for a Wheeled Robot”, 2002. [3]. Y. Koren, Senior Member, IEEE and J. Borenstein, “Potential Field Methods and Their Inherent Limitations for Mobile Robot Navigation”, Proceedings of the IEEE Conference on Robotics and Automation, Sacramento, California, April 7-12, 1991, pp. 1398-1404 [4]. Mauro Massari, Giovanni Giardini and Franco Bernelli-Zazzera Politecnico di Milano, “Autonomous Navigation System for Planetary Exploration Roverbased on Artificial Potential Fields” [5]. W.L.Xu, S.K. Tso, “Sensor-based fuzzy reactive navigation for a mobile robot through local target switching”, IEEE Transactions on Systems, Man and Cybernetics, Part C, Vol.29, No.3, 451-459, 1999. [6]. Xiaoyu Yang, Mehrdad Moallem, Rajni V. Patel, “A Layered Goal-Oriented Fuzzy Motion PlanningStrategy for Mobile Robot Navigation” IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, VOL. 35, NO. 6, DECEMBER 2005 [7]. K.M. Krishna, P.K. Kalra, “Perception and remembrance of the environment during real-time navigation of a mobile robot”, Robotics and Autonomous Systems, vol.37, 25-51, 2001. [8]. Anmin Zhu and Simon X. Yang, “A Fuzzy Logic Approach to Reactive Navigation of Behavior-based Mobile Robots”, Proceedings of the 2004 IEEE International Conference on Robotics 8 Automation New Orleans, LA Aprll2004 [9]. MENG WANG, JAMES N.K. LIU, ” FUZZY LOGIC BASED ROBOT PATH PLANNING IN UNKNOWN ENVIRONMENT”, Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, Guangzhou, 18-21 August 2005 [10]. L. A. Zadeh, \Fuzzy sets," Information and Control, vol. 8, pp. 338{353, 1965. [11]. H.G Tilleme Ed. , An overview of the mobile autonomous robot twente project MART 93, Univ. Twente, WA-315, 1993, pp. 315_319. [12]. Michael Young, “BEHAVIOUR BASED NAVIGATION OF NOMAD SUPER SCOUT ROBOT“,College of Sciences, Massey University Projects, Vol. 10, 2002 [13]. S. Thongchai, S. Suksakulchai, D. M. Wilkes, and N. Sarkar, “Sonar Behavior-Based Fuzzy Control for a Mobile Robot”, IEEE international conference on systems, Man, and Cybernetics, Nashville, Tennessee, October 8-11, 2000. [14]. Javid Taheri and Nasser Sadati, “A Fully Modular Online Controller for Robot Navigation in Static and Dynamic Environments”, Proceedings 2003 IEEE International Symposium on Computational Intelligence in Robotics and Automation July 16-20, 2003, Kobe, Japan. This essay is only for research purposes. If used, be sure to cite it properly! |
|
All images, coding, essays, and pages cannot be used without the prior
written consent of this web site. Copyright © 1996-2009. The Essay Depot. All rights reserved. Acceptable Use Policy | Copyright Inquiries | Privacy Policy |