Robotic Chain Formations ⋆ Paul M. Maxim * William M. Spears ** Diana F. Spears ** * Wyoming Department of Transportation, Cheyenne, WY 82009 USA (email: paul.maxim@dot.state.wy.us). ** Swarmotics LLC, Laramie, WY 82070 USA (email: wspears@swarmotics.com, dspears@swarmotics.com). Abstract: One important task in cooperative robotics is the self–organization of robotic chain formations in unknown environments. Surveillance and reconnaissance in sewers, ducts, tunnels, caves or narrow passageways, in general, are some of the applications for this type of formation. In Hettiarachchi et al. (2008) we provided simulation results for a novel chain formation algorithm that addresses this task. This paper presents the results of the chain formation algorithm implementation on five robots. Keywords: collaborative robots, robot localization, trilateration, chain formations. 1. INTRODUCTION One important task in cooperative robotics is the self– organization of robotic chain formations in unknown en- vironments. Surveillance and reconnaissance in sewers, ducts, tunnels, caves or narrow passageways, in general, are some of the applications for this type of formation. These applications are greatly augmented via the use of multi–robot teams. First, multiple robots can cover more territory than one. Second, if the multiple robot system is designed properly, there is no single point of failure. The failure of one or more robots will degrade performance, but the applications can still be partially accomplished. The scenario is as follows. A robot is placed at the entrance of the environment that is to be autonomously explored. This first robot will remain stationary and will wirelessly communicate information to a laptop. Other robots are released into the environment, by being placed in front of the first robot. The goal of the robots is to self– organize into a chain structure that reflects the shape of some portion of the environment. Because our localization technique combines localization and data communication, the robots can use this self–organized communication link to provide their positions and orientations in the environment back to the user. The user will see a diagram of the positions and orientations of the robots, relative to the position and orientation of the first robot, on the laptop screen. The real–time generated diagram informs the user of the shape of this portion of the environment. 1.1 Related Work For creating a chain formation algorithm, the behavior of social insects (e.g., ants) is one source of inspiration. For example, ants leave pheromones (a chemical) in the environment while they are foraging for food. This process of altering the environment is called “stigmergy.” One approach presented in Nouyan and Dorigo (2006) uses ⋆ This research has been financially supported by the Joint Ground Robotics Enterprise Program, United States Department of Defense. robotic chains as trail markers. Cyclic directional patterns are used to give the chains a directionality. The chain starts at a predetermined location (nest) and expands in a random direction until the goal location (prey) is reached. Other robots then navigate between nest and prey by following the robotic chain. The robots can only attach at the end of the chain and robustness was not tested. The environment used for experiments was rectangular in shape and had no obstacles. In a different approach, Mamei and Zambonelli (2005) make use of RFID tags that are deployed in the environment as digital pheromones. Robot formations such as columns, wedges, lines, and diamonds can be generated by assigning each robot a robot “friend,” as shown in Fredslund and Mataric (2002). Then each robot keeps its single friend at a desired angle θ. Depending on the desired shape of the formation, the robots are started out in the right order next to each other, with random headings. Robot formations are generated by chains of friendships and are able to handle smooth turns, but cannot deal with sharp turns. In Mead et al. (2007), control of robot formation shapes is achieved by treating each robot as a cell in a cellular automaton, where local interactions between robots result in a global organization. When the state of one of the robots (“initiator”) is changed, the change propagates to its neighbors. Although the algorithm is scalable, in the experiments the robots are initially placed in a formation and the robots start by detecting their neighbors. Exam- ples include changing the formation from a parabola to a line or from a parabola to a sine curve. Monteiro et al. (2004) use non–linear attractor dynamics to build controllers that implement formations. Robot formations are decoupled into pairs of leader–follower robots and each robot in the pair can be either a follower or a leader. These pairs can achieve column, oblique, and line formations. More complex formations can then be generated by combining these three basic formations.