Research On Self-organizing Cooperative Hunting By Swarm Robots Based On Simplified Virtual-force Model

In recent years, the swarm robotic system is becoming a rising research field. It is inspired by the complex natural systems, such as social insects(ants, bees, etc.) or groups of animals with collaboration. Its global behavior emerges from the local rules implemented on the level of individual robots. A very typical task platform for researching swarm robotic system is the swarm robotic hunting system, which is conducive to surrounding, searching and rescuing, group confrontation, formation keeping, collaborative transportation, targets defending and leaders guarding, etc. Being widely used in counter-terrorism, military, security and so on, the swarm robotic hunting system has important research value. However, there are many defects in former research in a swarm robotic hunting system, such as theoretical defects, relatively simple hunting environments, weak scalability and complicated multi-objective hunting algorithm, and so on. Aiming at these problems, swarm robots hunting based on artificial physical methods under complex environments is investigated in this dissertation. The main research contents are as follows:(1) Hunting by nonholonomic mobile swarm robots in unknown dynamic environments with convex obstacles. In this dissertation, we proposed a hunting theory and a hunting algorithm from the view of artificial physics. Unlike the hunting algorithm based on a loose preference rule, this hunting algorithm is based on a simplified virtual-force model. By reducing the necessary information needed to just position information of two nearest neighbors and the target, this algorithm simplifies and fastens the computing. Besides, the simple force functions in the simplified virtual-force model makes the hunting systems with more easily setting parameters possible. What’s more, the designed bionic intelligent obstacle avoidance function curve realized obstacles avoidance in unknown dynamic convex obstacles environments. Studies on the stability and the law of Leader emergence in the hunting systems show that, the hunting system is stable given a sufficient condition, and the leader emergence judgment conditions are that there are approximatively successive monotonic decays after a larger value in individuals’ force sequence. This is opposite with the Leader emergence judgment conditions in the hunting systems based on a loose preference rule. The reason for that is the virtual force from the two nearest neighbors is modeled as a repulsive force in the simplified virtual-force model. The farther distance between the leader and its two nearest neighbors, the smaller repulsion from its neighbors. That makes the Leader’s oscillation smaller in the process of Leader emergence.(2) Hunting by nonholonomic mobile swarm robots in unknown environments with non-convex obstacles. Based on the simplified virtual-force model, we proposed a control method for swarm robots following motions of barriers, which also just needs position information of two nearest neighbors and the target. In this method, the movement angle between following robots and obstacles increases the speed at the direction parallel to the obstacles. That ensures obstacles avoidance and obstacle quickly following at the same time. In addition, we analyzed the stability of the hunting systems theoretically and gave the conditions for successful following of non-convex static obstacles.(3) Hunting by nonholonomic mobile swarm robots in unknown environments with dynamic deforming obstacles. Based on the simplified virtual-force model, we proposed a control method for swarm robots following motions of dynamic deforming barriers, which also just needs position information of two nearest neighbors and the target. The movement angle set between following robots and obstacles, which gives consideration to both the vertical velocity and the parallel velocity with the direction of the obstacles’ side, can ensure rapidly following the dynamic deformation obstacles and maintaining a suitable distance at the same time. The algorithm of obstacles following can be applied to robots, non-convex dynamic non-deforming obstacles, non-convex static obstacles, dynamic or static convex obstacles. In addition, the stability of the hunting systems is analyzed in theory. The conditions for successful following of dynamic deforming obstacles and the characteristics of the hunting formation for increasing the number of robots are also given.(4) A cooperative multilayer hunting by nonholonomic mobile swarm robots in unknown dynamic complex environments with non-convex obstacles. By changing the force computation on the single layer to any layer on the hunting circles for robots in the simplified virtual-force model, and conducting the movement for robots among layers with the position information of their two nearest neighbors in the inner layer or the same layer, the cooperative multilayer hunting algorithm we designed endows the hunting system with high scalability, high reliability and strong obstacle avoidance ability. The stability of the multilayer hunting systems is also analyzed in theory.(5) A cooperative multitarget hunting by nonholonomic mobile swarm robots in unknown dynamic complex environments with convex obstacles. We change the force computation from a single target to any target in a multitarget hunting for robots in the simplified virtual-force model. The hunting target for each robot is determined by the multitarget task allocation algorithm, which only needs tasks information of two nearest neighbors in the field of 180 o facing the direction of the multitarget center. The proposed method needs little information and it is distributed, simple and efficient. The stability of the multitarget hunting systems is analyzed in theory. And the requirements for robot number within the group in multitarget task allocation are also presented.(6) The experimental platform---- SRf H(Swarm Robots for Hunting) is designed. The correctness and validity of the proposed basic hunting theories and algorithms is verified by this platform. Using the LINK UWB positioning system, the hunting experiments based on the simplified virtual-force model are carried out in environments both with convex obstacles and with no obstacles. The results show that the proposed hunting theories and the algorithms have good robustness, scalability and flexibility. Furthermore, the law of Leader emergence is further tested in practice. Because of the complexity of communication environments, coupled with the positioning error of positioning system itself, it is more difficult to analyze the individuals’ force from their two nearest neighbors. Therefore, in addition to judging by filtering occasionally, we need to change the judgment conditions to having a force sequence coming from the left or the right two nearest neighbors of the individuals. Since the actual situation often appearing monotonic sequence but not damped, there is no longer need the limit of monotonic decay. That is the difference between the practice and the ideal simulation.