To flock or not to flock: The pros and cons of flocking in long-range “migration” of mobile robot swarms Fatih Gökçe ErolSahin ¸ KOVAN Research Lab. KOVAN Research Lab. Dept. of Computer Eng. Dept. of Computer Eng. Middle East Technical Univ. Middle East Technical Univ. Ankara, Turkey Ankara, Turkey [email protected] [email protected] ABSTRACT vironmental cues [12, 14, 21]) to determine the direction of This study investigates the pros and cons of flocking in long- their travel. Among other aspects of the migration behav- range “migration” of mobile robot swarms under the influ- ior, the accuracy of the flocks to reach the very same feed- ence of different factors. We present a flocking behavior con- ing or breeding grounds has attracted much interest. In [3], sisting of three simple behaviors: heading alignment, proxi- Bergman and Donner first suggested that the flock migration mal control, and alignment to the desired homing direction. “increases the accuracy of orientation mechanism” which is The behavior drives a flock of robots from one location to known as the many wrongs principle. They suggested that another by sensing the magnetic field of the Earth. We pro- flocking suppresses the tendencies of the individuals to mi- pose that four factors influence the accuracy of reaching a grate in slightly different directions. Hence the flock can particular location with the proposed behavior; namely, av- align to an average direction of the preferences of the indi- eraging through the heading alignment behavior, the noise viduals giving a more accurate direction when compared to in sensing the homing direction, the differences in the char- the case of individuals. acteristics of the individuals, and the disturbances caused by Hamilton [8] and Wallraff [20] reiterated the many wrongs proximal control behavior. In a series of systematic experi- principle in their theoretical studies. Hamilton [8] suggested ments conducted with both physical and simulated robots, that “the orientation of groups of animals is more accurate we evaluate the effects of these factors in the accuracy of than that of individuals”. Assuming that (1) the spatial long-range “migrations” of flocks. goal is the same for all individuals, (2) the inaccuracies are represented by the deviation of individuals from the goal and (3) the individuals adopt their orientation to the mean Categories and Subject Descriptors direction of the individuals in the flock, he drew a series of I.2.9 [Artificial Intelligence]: Robotics theoretical curves with respect to flock size showing that av- erage deviation from the goal decreases with the flock size. General Terms Wallraff [20] suggested some methods to analyze the obser- vational data to investigate the effect of flocking to the accu- Algorithms, Performance, Design, Experimentation racy of orientation toward the goal direction and described their statistical implications. Keywords In [13], Rabøl et al. observed skylark flocks of different swarm robotics, flocking, migration sizes (1, 2, 3-5, and 6 or more) on their spring migration. They showed that the dispersion of migratory directions be- 1. INTRODUCTION comes less scattered with the size of flock. Later, Tamm [17] observed similar results by testing the hypothesis on homing Every year, certain animal and insect species flock to- pigeons with three to six flocks. By selecting flocks in a ran- gether to make long-range migrations to reach their feed- dom fashion, he showed that the flocks are more accurate ing or breeding grounds. Migration is an impressive phe- than individuals and their homing time is shorter than that nomenon because of its three important properties: (1) Very of individuals. However, some contradictory observations long distances scaling up to several thousands of kilometers were also reported. In [10], Keeton compared mean bear- are travelled. (2) Migratory animals and insects typically ings of single pigeons with that of flocks of four pigeons. He migrate in flocks (which may include millions) rather than reported no significant difference between single birds and as individuals. (3) Migration occurs in an accurate way de- flocks in terms of accuracy. In [2], Benvenuti et al. compared spite different environmental conditions and hazards. the orientation behavior of single birds with that of small Biological studies indicated that these animals mainly use flocks ranging from three to ten birds. Their results showed the magnetic field of the Earth [1, 21] (among various en- that small flocks do not orient more accurately than single CiteCite as: as: ToTo Flock flock or not toto flock:Flock: The The pros Pros and and cons Cons of flockingOf Flocking in long- in birds. In [6], Guilford et al. released pairs of homing pigeons rangeLong-Range “migration” “Migration” of mobile of robotMobile swarms, Robot FatihSwarms, Gökçe Fatih and Gökçe, ErolSahin, ¸ Erol in which none, one or both of the birds had previously been Proc. of 8th Int. Conf. on Autonomous Agents and Multia- Şahin, Proc. of 8th Int. Conf. on Autonomous Agents and Multiagent Sys- trained. They investigated whether unexperienced birds ex- gent Systems (AAMAS 2009), Decker, Sichman, Sierra and Castel- franchitems (AAMAS (eds.), May, 2009), 10–15, Decker, 2009, Sichman, Budapest, Sierra Hungary, and Castelfranchi pp. XXX-XXX. (eds.), ploit the knowledge of other bird to achieve a navigational CopyrightMay, 10–15,c 2009, Budapest, International Hungary, Foundation pp. 65–72 for Autonomous Agents and advantage or not. They found that unexperienced birds do MultiagentCopyright Systems© 2009, (www.ifaamas.org).International Foundation All rights for Autonomous reserved. Agents and Multiagent Systems (www.ifaamas.org), All rights reserved. 65 AAMAS 2009 • 8th International Conference on Autonomous Agents and Multiagent Systems • 10–15 May, 2009 • Budapest, Hungary not prefer to home together with their pairs. Recently, Simons has brought the almost forgotten many wrongs principle to light as a null model and general frame- vL work for testing the advantage of group navigation empiri- ω cally [16]. Taking the principle in its simplest form in which θ vR there are no characteristic differences between individuals 2 3 1 and contribution of individuals to the direction of flock are 4 0 equal, he showed that large group size increases the accu- 57 racy of group navigation. He emphasized that the principle 6 can be generalized to more complex scenarios in which there (a) (b) (c) are differences between individuals and the individuals con- tribute to flock direction in an unequal manner. Figure 1: (a) The Kobot robot platform. (b) The The work of Simons has rejuvenated attention to the many scaled drawing of Kobot illustrating the circular wrongs principle. Codling et al. studied the principle in a body, wheels, placement of the sensors and range scenario resembling to the migration of animals [4]. They for 2nd sensor. The sensors are placed uniformly at developed a point-mass movement model incorporating a 45◦ intervals. Each square patch in the gray scale biased random walk behavior and the group interactions. blob indicates the output of the sensor. (c) The They investigated the effect of navigational error, group size, body-fixed reference frame of robot is depicted. It interaction radius size and environmental turbulence to the is fixed to the center of the robot. The x-axis of the performance of the behavior to navigate a group from one body-fixed reference frame coincides with the rota- location to another. They found out that, with the excep- tion axis of the wheels. The forward velocity (u) tion of the high environmental turbulence case, the group is along with the y-axis of the body-fixed reference movement provides a navigational advantage. frame. The angular velocity of the robot is denoted In robotics, Guti´errez et al. [7] proposed a fully-distributed with ω. vR and vL are the velocities of the right strategy for the improvement of odometry in collective robo- and left motors, respectively. θ, current heading or tics. In this strategy, the robots improve their estimate of the robot, is the angle between the y-axis of the location by exploiting the estimations of their neighbors. body-fixed reference frame and the sensed North di- The estimate of each robot is associated with a confidence rection (ns). l is the distance between the wheels. level decreasing with the distance travelled by the corre- sponding robot. Each robot combines its own estimate and the received estimates of its neighbors using the confidence 2.1 Kobot mobile robot platform level of each estimate to get a more precise location infor- Kobot is a CD-sized, differentially driven and power effi- mation. They evaluated their strategy in simulations on a cient platform weighing only 350 gr with batteries. It has 8 foraging task in which the duty of the robots was to bring infrared (IR) sensors capable of kin and obstacle detection items from a resource site to a central place. Their results and a digital compass. The communication among robots showed that as the group size increases both the quality of as well as between the robots and a console is carried out the individuals’ estimates and the performance of the group through an IEEE 802.15.4/ZigBee compliant wireless com- improve. munication module. As reviewed above, the interest in the role of flocking in The infrared short-range sensing system (IRSS) measures long-range migrations has produced a number of hypotheses the range and bearing of kin-robots and other objects in and models in biological systems. Despite the results ob- close proximity. It consists of eight infrared (IR) sensors tained in simulations, coupled with few, sometimes contra- ◦ placed uniformly at 45 intervals, as shown in Figure 1(b) dictory observational data from animal flocks, the problem and a coordinator microcontroller that controls the sensors.