National University of Singapore

Transcription

1 National University of Singapore Department of Electrical and Computer Engineering EE4306 Distributed Autonomous obotic Systems 1. Objectives Equipment Preparation Introduction Distributed Autonomous obotic Systems (DAS) The MiroSot system ehavior based control of robots Fuzzy sets, fuzzy logic and fuzzy control The experiment Task 1: Forward motion to the ball Task 2: Obstacle avoidance Task 3: Goal Scoring Forward Motion Obstacle Avoidance Goal Scoring Assignment (OPTIONA): Designing a complete behavior-based controller for the MiroSot system eferences and further reading Objectives 1. Understand robot navigation in a micro-robot soccer (MiroSot) system. 2. Observe how fuzzy logic can be used to implement basic behaviors of forward motion and obstacle avoidance. 3. (Assignment) Design a behavior-based fuzzy control system that will autonomously control a MiroSot system. 2. Equipment 1. Personal Computer(s) with frame-grabber(s) (ideo Capture Card). 2. Micro-obot(s) working within a robotic soccer platform. 3. Graphical User Interface (GUI isual C++). 3. Preparation ead up on fuzzy-logic, behavior-based robotics and autonomous systems (see the list of references at the end of this manual. A references are available online). ead this manual thoroughly to get an idea of what to expect in the experiment. 4. Introduction Controlling multi-robot systems continues to be a challenging problem and an active research area. In this experiment, you ll gain first hand experience in controlling a multi-robot soccer system using a behavior-based strategy. Three independent behaviors, viz., the forward motion behavior, the obstacle avoidance behavior and the goal scoring behavior are implemented using

2 fuzzy logic modules. You ll observe how fuzzy rules and the membership functions affect the functioning of the robots. In the assignment, you ll be required to identify and implement other relevant behaviors and coordinating strategies between the behaviors. Though the assignment is optional, it is strongly recommended that you do it to get a fuller appreciation of the subject Distributed Autonomous obotic Systems (DAS) DAS are interesting due to several reasons: Tasks may be inherently too complex (or impossible) for a single robot to accomplish. The efficiency of accomplishing a task may be vastly improved by using robots that act in parallel. uilding and using several simple robots can be a cheaper, simpler and more fault tolerant alternative to using a single, sophisticated robot. The cooperative behaviors developed in DAS systems could give useful insights into the behaviors of biological communities like ants and bees. The beginnings of distributed robotics can be traced to the late 1980s, when researchers began investigating issues in multiple mobile robotic systems. Prior to this, research had mainly centered on either single-robot systems or distributed problem-solving systems that did not involve robotic components. The field of distributed robotics is still new enough that no single area within this domain can be considered mature. The fundamental question in DAS is, Given a task and an environment, how can a team of mobile autonomous robots be designed and coordinated so that the task is accomplished? 4.2. The MiroSot system obot soccer systems are a publicly appealing but formidable challenge to DAS researchers. obot soccer researchers aim to eventually create robotic teams that exhibit cooperative behaviors and morphologies sophisticated enough to challenge human soccer teams. They form a platform on which approaches to DAS can be tested and the state of the art measured. An analogy would be IM s Deep lue that challenged and beat the human chess grandmaster, Gary Kasparov, with chess being chosen as a benchmark problem for Artificial Intelligence. The robots used in the setup are small in size (7.5-cm Cubic), fully/semi autonomous. The robot soccer environment involves multiple robots that need to collaborate in an adversarial environment to achieve specific objectives. The assigned color patches on top distinguish the robot teams. Wireless F communication is used to control to robot from a host computer. Figure 1: The robots in the MiroSot system.

3 The MiroSot system used in the experiment consists of the robots, the vision system, the controller and the communication system. The images of the robots on the playground are captured by an overhead camera and sent to a machine vision algorithm. The M algorithm determines the positions of the robots from the images, and this information is used by the controller to determine what the robots should do next. The controller generates commands to the robots in the form of right and left wheel velocities, and is sent to them through an F communication system. In your experiment, you ll be dealing with the controller which is implemented using fuzzy logic. However, please feel free to ask the lab demonstrators questions on any other parts of the system as well. ision System The vision system consists of a CCD camera, frame grabber and drivers. The field frames are captured every 20 milliseconds and displayed on the computer screen from which the positions of robots and the ball are recognized by software. The height of camera or sensor system is greater than or equal to 2m above the field. Kinematics of the Mobile obot elocity of point P (Figure 2.) is = ( r ω), a x where, a x is the unit vector along X-axis, velocity of the wheel. r is the radius of the wheel and ω is the angular Figure 2: The wheel velocities Posture of a mobile robot The posture P of a mobile robot is described with the position ( x, y) and the heading angle θ. x P = y ( x, y) : obot Position θ θ : obot Orientation

4 Figure 3: The robot posture Control input The left and right wheel velocities constitute the control inputs to the robot. I.e., = w v U, where, v is the linear velocity of the robot and ω is its angular velocity. The linear and angular velocities of the robot are related to the control input as follows: 2 ) ( ) ( v r r = = = = ω ω ω The Kinematic equation of the center of the bi-wheel mobile robot is:... c c c y x θ = w v sin 0 cos θ θ

5 Instantaneous center of rotation (IC): ( = 2 ( ) = ( + 2) + ) /( :adius of rotation When the robot moves along a straight-line path, the radius of rotation is infinity. The robot will move along the straight line with equal left and right wheel velocities. I.e., = Infinity =. When the robot rotates about itself, the radius of rotation becomes zero with equal wheel velocities in opposite directions. = 0 = 4.3. ehavior based control of robots ehavior-based robotic control is a robotic control paradigm developed in the late 80s. Intelligent robot control strategies can be categorized as deliberative, reactive or a hybrid of these two. In deliberative approaches, the intelligent robot controller uses a centralized world model for verifying sensory information and generating actions in the world. The information in the world model is used by a central planner to produce the most appropriate action in the real world. Though traditionally deliberative approaches have been prominent, towards the late 80s they were increasingly criticized for scaling poorly with the complexity of the environment and for using artificial symbols that, in themselves, have no meaning. Purely reactive controllers, on the other hand, maintain no internal models and perform no search. They maintain a collection of pre-programmed condition-action pairs with minimal internal states. Thus, every time the robot receives a stimulus, the corresponding action is simply performed (the robots react to the stimuli). Due to their direct coupling between sensing and action, the reactive systems have very short response times. Hybrid architectures seek to combine both the above approaches, usually by using reactive techniques for lower level control and deliberative approaches for higher-level planning. ehavior-based approaches can be considered as an extension of reactive architectures, though their computation isn t limited to look-up and execution of simple functional mappings. ehavior-based controllers consist of a collection of behaviors that achieve and/or maintain goals. For example, avoid-obstacles behavior maintains the goal of avoiding obstacles, and move-to-the-ball behavior maintains the goal of moving to the robot to ball. ehaviors are implemented as distinct modules using hardware or software. Each behavior can take inputs from the robot s sensors and/or other behaviors and send outputs to the robot s actuators or other behaviors. Thus, a behavior-based robotic controller is essentially a network of interacting )

6 behaviors. The two major concerns in designing behavior-based controllers are, 1) breaking down the robot s overall behavior into individual behavior modules, and 2) coordinating the outputs of the individual behavior modules so as to obtain an overall intelligent behavior. ehavior-based approaches have especially been found to be useful in control and coordination of autonomous multi-robot systems. In this experiment, the overall control system is assumed to be organized in a behavior-based manner. You ll get a feel of how a few individual behaviors ( forward-motion behavior, forward-motion-and-obstacle-avoidance behavior, goal-scoring behavior etc) are implemented using fuzzy logic. Your assignment is to complete the overall multi-robot behavior-based controller by adding in more behaviors and suggesting how to coordinate them Fuzzy sets, fuzzy logic and fuzzy control Fuzzy sets, as opposed to the conventional crisp sets, have members that have varying degrees of membership in the set. Fuzzy sets are ideal for numerically representing linguistic concepts, such as old or young. For example, a 40-year old person may belong to the fuzzy set Old with a membership degree of 0.6, while a 80-year old may have a membership degree of 0.9. Similarly, 35 degree centigrade may have a membership degree 0.7 in the set Warm. Membership degrees for a given fuzzy set can be obtained from its membership function, µ. Thus, µ ( 40) = 0. 6, and µ ( 80) = Old Old Fuzzy logic is a system of logic that reasons with vague concepts, and is built on fuzzy set theory. For example, IF the person is Old THEN give him High priority might be a fuzzy logic statement that might be present in, say, a hospital queuing system. Old and High are fuzzy linguistic variables, and the IF-THEN rule is calculated according to the fuzzy implication used. In a fuzzy control system, the control laws are expressed using fuzzy logic statements. For example, the fuzzy statement, IF robot is Near a resource, then move Fast towards it might be a fuzzy logic statement in a robot s fuzzy control system. The laws present in fuzzy control systems can be easily parameterized, and the parameters then tuned through artificial evolution so that a control system of acceptable competence is reached. 5. The experiment The experiment consists of three tasks and an optional assignment. In the first task, fuzzy logic is used to implement the forward-motion behavior module. In the second, an obstacle-avoidance behavior module is similarly implemented. Finally, the goalscoring behavior module is implemented using fuzzy-logic. You ll use the ideas you come across in this experiment to design a behavior-based autonomous controller for the MiroSot system in your assignment. Though the assignment is optional, it is strongly recommended that you do it to get a fuller appreciation of the subject Task 1: Forward motion to the ball The objective of the task is to make the robot move towards the ball on the playground. The two fuzzy set inputs are distance error φ and angle error ϕ of the robot with respect to the ball. The final outputs will be the left and right velocities of the wheels of the robot. The robot will move left when the left velocity is less than the right velocity and vice versa. The

7 D velocity difference between them will determine how fast the robot turn to the desired angle. Only one robot is used in this section. eft wheel velocity Positive angle error, ϕ obot all ight wheel velocity Distance error, φ Figure 4: The angle and the distance errors. Two fuzzy rule sets control the velocity difference ( ) between the two wheels and the base velocity ( ) respectively. If is positive (i.e., > ), the robot turns right and vice versa. D The base velocity is determined by a fuzzy controller that takes in the angle and the distance errors as inputs. Thus, the velocities involved are: D = ( ) / 2 = = + D D The rule-set determining the velocity difference ( ) and base velocity ( ) are as follows: D D Figure 5: The rule set for determining D

8 Figure 6: The rule set for determining The inputs for both the above rule sets are the distance error and the angle error, as shown in figure 7. Distance error, φ Angle error, θ Fuzzy rules determining D Fuzzy rules determining Calculate, Figure 7: Implementation of the forward-motion behavior module using fuzzy-rule sets Task 2: Obstacle avoidance The objective of the obstacle avoidance task is to avoid any obstacle along the robot s path to the ball on the playground. In addition to the fuzzy rules mentioned above, a 3 rd rule set is used to cause the robot to stay away from the obstacle if it is in its path to the ball. This rule set generates an additional velocity difference between the wheels,, using the orientation information of the obstacle, consisting of the distance error ( φ ) between the robot and the obstacle, and the angle error ( θ o ) between them. o o

9 φ θ obot θ o φ o Obstacle Figure 8: The angle and distance errors between the robot and the obstacle. The new velocity equations are thus, DO = = = D + + O DO DO The fundamental principle is that if the robot is running into an obstacle, then the 3 rd fuzzy rule set adds some velocity to one wheel and subtracts some from the other in such a way that the robot will turn away from the obstacle. Thus, the additional velocity difference might make the robot turn more to avoid the obstacle or turn less so that it does not run into the obstacle. The respective rules are, IF (obstacle is in front) AND (obstacle is quite far) THEN (turn slowly away) Y ( o = small value) IF (obstacle is at the back) O (obstacle is very far) THEN (follow original path to ball) Y ( o = 0) Etc. The complete rule-set for determining the O is as follows: O

10 Distance error, φ Figure 9: The rule set for determining O Fuzzy rules determining D Calculate, Angle error, θ Obstacle distance, φ o Obstacle angle, θ o Fuzzy rules determining Fuzzy rules determining o Figure 10: Implementation of the obstacle avoidance behavior module using fuzzy-rule sets. 5.3 Task 3: Goal Scoring The objective of this task is to kick the ball into the goal. In addition to the inputs listed in task 1 (forward-motion), viz., the angle error and the distance error between the robot and the ball, goal scoring includes the angle error between the robot and the goal.

11 Desired path to ball Desired final posture θ obot θ g Goal Figure 11: The robot aligns itself so that it forms a straight line with the ball and the goal. Two new rule-sets are introduced in addition to the two already present for forward motion. The third rule-set generates a velocity difference that causes the robot to approach the ball in such a manner that the robot, the ball and the goal-post form a straight-line, and caters for the case when the ball is between the robot and the goal along the x axis. The fourth rule-set generates when the robot is between the ball and the goal. The new velocity equations are thus. DO g = = = D + + g DO DO g The rule base for determining g is as follows:

12 Distance error, φ Figure 12: The rule-set for determining g Fuzzy rules determining D Calculate, Angle error, θ Fuzzy rules determining Goal angle, θ d Fuzzy rules determining g Figure 13: Implementation of the goal-scoring behavior module using fuzzy-rule sets. The behavior of all the above rule sets can be adjusted by adjusting the membership functions of the linguistic variables. The linguistic variables range from N (Negative ig) to P (Positive ig) for velocities, angle errors and distance errors. The angle errors vary from -180 to 180 degrees, while the distance errors vary from 0 to 180 cm. The interpretations of the various linguistic variables are shown in the following table: inguistc Angle Error Inputs Distance Error Inputs elocities ariable N ery near all at left and behind ery slow NM Near all at left side Slow NS Not so near all at left and in front Not so slow ZZ Okay dist all straight ahead Okay PS Not so far all at right and in front Not so fast PM Far all at right side Fast P ery far all at right and behind ery fast

13

14 5.4. Forward Motion Tuning the control system parameters in the robot 1. Place the robot and the ball on the field as shown below. The position of the robot can be anywhere, as long as it facilitates observation of the robot s movement towards the ball. oading the fuzzy control laws Figure Click the Open Fuzzy Console (see Figure 14) button to bring up the fuzzy console. This is where you view and adjust the fuzzy rule-sets that control the robot. 3. In the fuzzy console that pops up (Figure 15), click the oad button and load the fuzzy configuration file fuzzyconfig\section1a(forward).txt.

15 Figure Clicking the Graph button (Figure 15) brings up the following window (Figure 16) displaying the fuzzy rules and membership functions. You can use this to screen capture the rules for your report. To do this, simply click AT+PINTSCN to copy the window image to the clipboard. You can then open Paint or Word and paste the figure onto the file. Figure Click the eady button (Figure 17) to start the machine vision algorithm. 6. The controlling computer sends the control inputs to the robots in the form of right and left wheel velocities. The robots have an on-board PID controller making the wheel

16 motors track the velocity reference signal send by the computer. To tune this controller, set Kp=20, Kd=7 and Ki=0 (the PID parameters in the robot s controller) in their respective check-boxes (Figure 17). You could try different values for these parameters and explain the effect on the robot s behavior in your report. ecording the robot motion Figure The robot s motion needs to be captured so that you can use it in your reports. To do this, select Debug Dialog from the iew menu (Figure 18). This will bring up the debug window that will allow you to record the robot s path.

17 iew menu Debug Window Figure 18 Forward motion In the following steps, the fuzzy control laws loaded will be used to guide the robot to the ball. The strategy is to feed back the error between the position of the ball and the robot and use it to generate the actuating signals. 8. Make sure you ve selected the Forward Motion radio button. Click the Start button to start moving the robot to the ball. You can stop the robot by clicking the Stop button (Figure 19).

18 Figure 19 Screen Capture 9. Click the Stop ideo-button, and move the recorder slider-bar (Figure 20) to get to the screen you want to capture. Capture the screen and paste it to a Paint or Word file. Figure 20

19 Tuning the fuzzy membership functions 9. Tune the fuzzy membership functions to get the robot to trace a smoother path in the fuzzy membership console. To do this, select Triangle from the Shape of Function drop-list. You can then adjust the membership functions of the linguistic variables by simply typing the new values in the respective text boxes. 10. Note that in this section, there are two fuzzy rule sets that can be tuned. These can be selected for tuning using the Type of Application Usage drop-list. Selecting 1 from this drop-list allows you to tune the fuzzy rule-set that generates, and 2 allows you to D tune the one generating, The modified membership function epetition using different fuzzy rule-sets Figure epeat the experiment using the stored fuzzy-rule sets fuzzytconfig\section1b(gaussian).txt and fuzzyconfig\section1c(reduce usage1 output).txt. You must try to get some feel of how the robot s behavior varies with different membership functions and starting ball/robot postures. Comment on your observations in your report. What are some of the advantages and disadvantages in using fuzzy logic to implement robot behaviors?

20 5.5. Obstacle Avoidance. In this section, fuzzy control laws will be used to implement the obstacle avoidance behavior. 1. oad the fuzzy configuration file fuzzyconfig\section2a (obstacle).txt 2. Place a second robot between the original robot and the ball to serve as an obstacle. Check the Obstacle checkbox to track this obstacle, and the Obstacle avoidance radio button to specify the behavior as shown above. Click the eady button to start the machine vision algorithm. Figure On the menu iew, select Debug Dialog to start recording the robot motion. 4. Click Start to run the loaded fuzzy rules. The robot should try to avoid the obstacle and reach the ball. 5. Tune the fuzzy rules to get a better obstacle avoidance performance. Explain your observations Goal Scoring In this section, fuzzy logic will be used to implement the goal-scoring behavior. 1. oad the fuzzy configuration file, fuzzyconfig\section3a (kickslow). 2. Place the robot as shown in figure 23.

21 3. Uncheck the Obstacle checkbox. Select the Scoring radio-button. Click eady. Figure From the iew menu, select the Debug Dialog to start recording. 5. Click Start to run the fuzzy controller. 6. Try tuning the controller to get a better performance. 6. Assignment (OPTIONA): Designing a complete behavior-based controller for the MiroSot system The aim of this assignment is to give you a better idea of how to design a complete autonomous behavior-based controller for a multi-robot system. Your assignment consists of two portions: 1. Complete the set of behaviors you think the robots must have. 2. Come up with a strategy to coordinate the outputs from the behavior modules. Please keep in mind the following points. 1. Use fuzzy logic and/or any other algorithm you may consider necessary to implement the behaviors and their coordination. 2. Each behavior module has access to the following data computed by the system s machine vision algorithm.

22 a. The positions, velocities and orientations of the player and opponent robots. The velocity is a magnitude that has the same direction as the robot s orientation. b. The position and the velocity of the ball. The velocity information consists of its orientation and magnitude. c. All information related to the playground (the borders, positions of the ball, etc.) 3. There are three robots in a team: the goalie, the striker and the defender, and you may choose to design behaviors for each of these (thus, for example, the goalie may never leaves the goal-post). Or you could choose the robots to have access to A behaviors (the striker comes to defend the goal-post when an opponent is attacking). 4. The final outputs are the right and left wheel-velocities of the robots. ut you may choose to have intermediate behaviors giving output to other behavior modules. 5. Your description needn t contain all the details for implementation. You ll, however, need to give enough information as is necessary to convey that you have a very clear idea on how to implement the controller. If you re using fuzzy logic, list clearly the linguistic variables, the inputs, the outputs and a few representative rules that you use. 6. Provide a diagram to show how your behaviors are arranged and interact with each other. 7. eferences and further reading 1. Federation of International obot-soccer Association ( 2. Maja J Mataric, "ehavior-ased obotics", in the MIT Encyclopedia of Cognitive Sciences, obert A. Wilson and Frank C. Keil, eds., MIT Press, April 1999, ( 3. Maja J Mataric, " ehavior-ased Control: Examples from Navigation, earning, and Group ehavior", Journal of Experimental and Theoretical Artificial Intelligence, special issue on Software Architectures for Physical Agents, 9(2-3), H. Hexmoor, I. Horswill, and D. Kortenkamp, eds., 1997, rooks,. A. "A obust ayered Control System for a Mobile obot", IEEE Journal of obotics and Automation, ol. 2, No. 1, March 1986, pp ; also MIT AI Memo 864, September ( 5. Fuzzy-ogic jump start (