Fuzzy Logic for Behaviour Co-ordination and Multi-Agent Formation in RoboCup Hakan Duman and Huosheng Hu Department of Computer Science University of Essex Wivenhoe Park, Colchester CO4 3SQ United Kingdom {hduman, hhu}@essex.ac.uk Abstract. Robots participating in a soccer game need to determine the position of the ball, other robots, and the goal positions using real time visual tracking, along with being able to navigate safely, move the ball towards the opponents goal, and co-operate with teammates. Each soccer robot is equipped with basic behaviours such as chasing the ball and shoot it at the goal. Although the single-agent behaviours are very important, the issue of co-operation, or formation, among multiple agents in such a domain is essential. In this paper, we discuss the importance of robot formation in RoboCup and introduce new reactive behaviours and their co-ordination, based on Fuzzy Logic Control, to achieve cooperation among the soccer playing robots. 1 Introduction The Robot World Cup, RoboCup, is an international competition designed to encourage artificial intelligence (AI) and intelligent robotics research by providing a standard problem, a soccer game, where a wide range of technologies can be integrated [Kitano et al., 1997]. Designing a robot to play football is very challenging because the robot must incorporate the design principles of autonomous agents, multi-agent collaboration, strategy acquisition, real-time reasoning, strategic decision making, intelligent robot control, and machine learning. In the middle-sized RoboCup league, robots are playing in a 8.22m x 4.57m green-floor area surrounded by walls. The ball is an official FIFA size-4 soccer ball and the size of goal is 150x50cm 1. In a game there are two competing teams. Each team can have up to four robot players with size less than 50cm in diameter. For playing an aesthetic and effective game of robotic soccer each soccer agent must be equipped with a set of basic behaviours [Hu et al., 1999]. 1 RoboCup Middle-Size League Regulations and Rules: http://www.robocup.org
Depending on whether the agent fills the role of the goal keeper or of a field player, there are different basic skills. The goalie is very simple minded and just tries to keep the ball from rolling into the own goal. The field players have a much more elaborate set of basic behaviours. These basic behaviours have been adopted from the CS Freiburg Team [Gutmann et al., 1999]. The first three skills below concern situations when the ball cannot be played directly, while the two last skills address ball handling: Search-ball: Turn the robot in order to find the ball. If the ball is not found after one revolution go to home position and search again from there. Observe-ball: Set the robots heading such that the ball is in the center of focus. Track the ball without approaching it. Go-to-position: Plan and constantly re-plan a collision free path from the robot s current position to a target position and follow this path until the target position is reached. Move-ball: Determine a straight line to the goal which has the largest distance to any object on the field. Follow line at increasing velocity and redetermine the line whenever appropriate. Shoot-ball: To accelerate the ball either turn the robot rapidly with the ball between the flippers or use the kicker-mechanism. The decision on which mechanisms to use is made according to the current game situation. Although the single-agent behaviours are very important, the issue of cooperation among multiple agents in such a domain is essential. It should be emphasised that co-operation in our approach is described as a formation-based attack or defence for the soccer playing robots. Formations allows individual team members to concentrate their sensors across a portion of the environment, while their partners cover the rest [Balch et al., 1998]. As in a real soccer match, formation-based attacks would increase the effectiveness and performance of tactical teamwork when combining with ball manipulation behaviours. In this paper, we introduce reactive behaviours, based on Fuzzy Logic Control, to achieve observation-based formation among the soccer playing robots. Although many of the participating teams are designing their teams based on the multi-agent co-operation, most of them employ formation control using explicit communication. Other approaches to similar tasks include [Arkin, 1998], [Parker, 1999], [Pirjanian et al., 2000]. The introduced method has been inspired by [Balch et al., 1998], which achieves formation among multiple robots by using a referenced-based (e.g. leader, neighbor, center) type of cooperation. The coordination of behaviours consisting of a hierarchical behaviour-based fuzzy control architecture adapted from [Duman et al., 2000] is introduced in section 3. Section 4 presents experimental results based on the proposed method. Finally a brief conclusion is given in section 5.
2 Fuzzy Control and Multi-Agent Formation 2.1 Fuzzy Logic Control Fuzzy control is one of the more active areas of application of fuzzy logic and the underlying fuzzy set theory introduced by [Zadeh, 1973]. A fuzzy logic controller works by encoding an experts knowledge into a set of rules which are smoothly interpolated and the resultant is defuzzified to give a crisp output. Using fuzzy logic techniques provide control, which is less sensitive to sensor errors since information is always assumed to be imprecise. Consequently such fuzzy logic control techniques are used for several aspects of autonomous robot control such as wall following and targeting a specified goal position [Saffiotti, 1997]. The autonomous soccer robots, which consist of different simple fuzzy behaviours, should deal with very different control situations, e.g. follow a teammate by holding constant distance while avoiding an obstacle. 2.2 Multi-Agent Formation Several formations exist for a team of four robots. In [Balch et al., 1998], the following are considered: Line, where the robots travel line-abreast. Column, where the robots travel on after the other. Diamond, where the robots travel in a diamond. Wedge, where the robots travel in a V or. Despite of the fact, that only four players are playing in each team, and obviously one is the goalkeeper, there are only three robots, which can be used to employ the offensive and defensive formations. [Balch et al., 1998] also identifies three different techniques for formation position determination: Unit-centerreferenced, Leader-referenced, and Neighbor-referenced. The most appropriate referenced-based type of cooperation for our soccer robots is the Leaderreferenced. The robot which dribbles the ball towards the oppents goal is being considered as the leader, while others stay on the side ready to recover the ball from any loses or pass the ball to the next closest teammate. The task of each robot is to simultaneously move to a goal location, avoid obstacles with other robots and maintain a formation position. In case of a defence, the robots organization choice can differ between the diamond and wedge formation. The diamond formation is used at game start and when all team robots are in the same side of the field. Whereby the wegde formation V has the advantage to prevent the opponent from dodging the defence and garanties a higher chance of stealing the ball from the opponents player. Figure 1 (a) and (b) illustrates above configurations. The following section describes experimental results for group formation of multiple robots.
Attacker Left Defender Right Leader Left Right Goalkeeper Figure 1: (a) Diamond formation at game start and defence. (b) Wedge formation during attack to the opponents goal. 3 The methodological approach 3.1 Robot platform The introduced system is implemented on a Pioneer 1 and Pioneer 2 mobile robot. The Pioneer 2 is equipped with an overhead vision system connected to a Ellips Rio frame grabber, 8 ultrasonic sensors, 2 position encoders. The Pioneer 1 basically consists of 7 ultrasonic sensors and 2 position encoders. The front sonars on both of the robots are used for obstacle avoidance and following a target. 3.2 Individual Behaviours The autonomous soccer robots, which consist of different simple fuzzy behaviours, should deal with very different control situations such as follow a teammate or ball while avoiding an obstacle. In this section we describe these elementary behaviours that are essential for our approach. These behaviours consist of 1) obstacle avoidance 2) target following and 3) velocity control. Every behaviour is specified using a fuzzy rule-base and generate an output by fuzzy inferencing.
Obstacle Avoidance In the robotic soccer field, there are often obstacles between the robot and its goal position. These can be in form of a wall or opponent player. Different possibilities can be taken to avoid obstacles. We have adopted this behaviour from [Duman et al., 2000]. It controls the heading of the robots based on the sonar readings to avoid collisions with obstacles and maintain a safe trajectory. Target Following The target following behaviour consits of two main tasks. The leader of the group follows the ball whereby the rest of the robots are following the leader. This behaviour controls the heading of the robot based on the input from vision system. The difference between the target position and its origin generate the vector d. Depending on the role of the robot in the group, it either follows the ball or the team leader. Figure 2 shows the target following behaviour for a robot, following a leader, that tries to maintain its formation position. Heading 2 1 d Figure 2: Target Following Behaviour Velocity Control Besides the target following behaviour, the formation approach requires multiple robots to form up and move in a specified distance to its neighbours. The velocity control behaviour controls the speed of the robot based on the sonar readings. Depending on how far out of position the robot is, the speed of the robot is being increased or decreased (Figure 3). Three designated areas a, b and c are considered for the robot. a) the robot is too close to its neighbour; the desired velocity is set to be decreased b) the robot is within the tolerance area; the velocity is kept constantly c) the robot is far out of position; the desired velocity is set to be increased Velocity slow constant fast -20-10 0 10 20 a b c Figure 3: Velocity Control Behaviour
3.3 Hierarchical Behaviour Co-ordination When a soccer playing mobile robot operates, several elementary behaviours of different type and goal can be active at the same time. This interaction can take the form of behavioural co-operation or competition. Each behaviour can be tuned independently to be more effective in its own context. In this way a complex behaviour can be obtained base on simpler behaviours. The fuzzy hierarchical controller, as shown in Figure 4, has been inspired by the technique proposed by Saffiotti [Saffiotti, 1997] for blending multiple behaviours with different tasks. Obstacle Avoidance Target Following Metarules Defuzzification Output: Heading Figure 4: The hierarchical Behaviour Co-ordination As described before, fuzzy logic was used to implement basic behaviours of the robot, e.g. obstacle avoidance and target following. Moreover the fuzzy metarules are implemented to describe strategies of behaviour arbitration. Depending on sensor readings the output control variable can be derived. The discounted values are then merged together according to a set of context rules of the form if condition/context then behaviour, telling which behaviour should be active in each situation and to what degree. Finally, the resulting blended function is defuzzified to produce a crisp control. For instance, in case an obstacle being detected, the obstacle avoidance behaviour weight needs to be increased respectively, at the expense of the other concurrent behaviours. When the obstacle is only partially close, both behaviours are partially activated.
4 Experimental Results In the experimental setup we used two Pioneer robots (see section 4.1 for description) for achieving multi-agent formation and co-operation. In order to monitor the formation position and orientation of the robots, the experimental runs were conducted in a test area measuring approximately 6 by 4 metres with an overhead camera tracking system. The robots were directed to navigate to the opponents goal area while avoiding robots from the other team, here in form of a pillar. The robot that has the ball is assigned to be the team leader. The other robot's task is to maintain its position in the group formation. Figure 5 presents the formation and co-operative behaviour from the viewpoint of a lab camera as well as from a vision system mounted on top of the robot. Figure 5: Formation and co-operative behaviour. Team formation and strategic attack to the opponents goal. Top row: Lab camera. Bottom row: An vision system mounted on top of the robot The results show that the robot, which is on lead searches for both, ball and opponents goal. The leader approaches the ball and starts dribbling it to the opponents goal area. Since the view area of the overhead camera is narrow, it is easy to lose the target. Hence, the robot searches the leader constantly. The leading robot detects an obstacle and starts avoiding it. We are assuming that the obstacle here is an opponents player that tries to block our robots path. The leader starts turning away from the obstacle. The opponents player follows the leader of our team and tries to block again. Now, the formation is used for tactical attack. The robot that has the ball loose its position in the formation and start a cooperative behaviour by passing the ball to the other robot, which has an empty path to the opponents goal.
5 Conclusions In this paper we have presented an approach to group formation for multiple autonomous mobile robots in general, soccer robots in particular. Elementary behaviours are implemented through fuzzy rules which provide robust and smooth navigation capabilities for a mobile robot. The generation of complex behaviours by combination of simpler behaviours has been proven to be effective and advantageous. Currently, our approach has been implemented on two robots. The first robot, a Pioneer 1 robot, serves as a leader of the group and the second robot, a Pioneer 2 with an overhead camera, tries to maintain the formation. Experimental results confirm that the approach produces acceptable co-operation by organising the robots in formation. We intend to employ this approach to a team of four soccer playing robots. References [Arkin, 1998] Ronald C. Arkin. Behavior-based Robotics. Inteligent Robotics and Automous Agents series. MIT Press, May 1998. [Balch et al., 1998] Tucker Balch and Ronald C. Arkin. Behavior-based Formation Control for Multi-Robot Teams. IEEE Transactions on Robotics and Automation, Vol. 14, No. 6, December 1998. [Duman et al., 2000] Hakan Duman and Huosheng Hu. Hierarchical Fuzzy Behaviour Coordination for Reactive Control of an Autonomous Mobile Robot in RoboCup. EUREL Robotics 2000, Salford, 12-14 April 2000. [Gutmann et al., 1999] J.-S. Gutmann, W. Hatzack, I. Herrmann, B. Nebel, F. Rittinger, A. Topor, and T. Weigel, The CS Freiburg Team: Playing Robotic Soccer Based on an Explicit World Model, The AI Magazine, June 1999. [Hu et al., 1999] Huosheng Hu, Kostas Kostiadis and Zhenyu Liu. Coordination and Learning in a team of Soccer Robots, Proceedings of the IASTED Robotics and Automation Conference, Santa Barbara, CA, USA, 28-30 October 1999. [Kitano et al., 1997] Kitano H., Tambe M., Stone P., Veloso M., Coradeschi S., Osawa E., Matsubara H., Noda I., and Asada M. The RoboCup Synthetic Agent Challenge, International Joint Conference on Artificial Intelligence (IJCAI97), 1997. [Parker, 1999] Lynne P. Parker. Cooperative robotics for multi-target observation. Intelligent Automation and Soft Computing, special issue on Robotics Research at Oak Ridge National Laboratory, 5(1):5-19, 1999. [Pirjanian et al., 2000] Paolo Pirjanian and Maja Mataric. Multi-robot Target Aquisition using Multiple Objective Behavior Coordination. IEEE International Conference on Robotics and Automation, San Francisco, April 2000. [Saffiotti, 1997] Alessandro Saffiotti. The Uses of Fuzzy Logic in Autonomous Robot Navigation: a catalogue raisonné. Soft Computing 1(4):180-197, 1997. [Zadeh, 1973] Lotfi Zadeh, Outline of a new approach to the analysis of complex systems and decision processes. IEEE Transactions on Systems, Man and Cybernetics, Vol. 3, No. 1, pp. 28 44, 1973.