本文選題：群機器人 + 簡化虛擬受力模型��； 參考：《湖南大學》2016年博士論文

【摘要】：群機器人系統(tǒng)研究是最近興起的一個熱點,它受啟發(fā)于復雜的自然系統(tǒng),比如社會性昆蟲(螞蟻、蜜蜂等)或者有協(xié)作的動物群體,群機器人系統(tǒng)的全局行為涌現(xiàn)自個體機器人層面上實施的局部規(guī)則。群機器人圍捕系統(tǒng)是群機器人系統(tǒng)研究的一個非常典型的任務平臺,具有重要研究價值,有利于實現(xiàn)包圍、搜救、群體對抗、隊形保持、協(xié)作搬運、目標保衛(wèi)以及領導護衛(wèi)等,可廣泛用于反恐、軍事、安全保衛(wèi)與警戒等方面。然而目前的群機器人圍捕系統(tǒng)理論尚不完備,所涉及的圍捕環(huán)境相對簡單,圍捕系統(tǒng)的可擴展性不強,多目標圍捕算法復雜。針對上述問題,本文研究基于人工物理方法的復雜環(huán)境下群機器人圍捕系統(tǒng),主要研究內(nèi)容如下:(1)未知動態(tài)凸障礙環(huán)境中非完整移動群機器人圍捕。不同于基于松散偏好規(guī)則的圍捕算法,而是從人工物理角度出發(fā),提出了基于簡化虛擬受力模型的圍捕理論和算法,該算法只需要兩最近鄰和目標的位置信息,算法簡單高效;簡化虛擬受力模型中簡單的受力函數(shù)實現(xiàn)了參數(shù)較易設置的圍捕系統(tǒng);設計的仿生智能避障函數(shù),實現(xiàn)了未知動態(tài)凸障礙物環(huán)境下的良好避障。在理論上分析了圍捕系統(tǒng)的穩(wěn)定性,研究了Leader涌現(xiàn)的規(guī)律。Leader涌現(xiàn)的判斷條件是受力出現(xiàn)一個較大值后逐步近似單調衰減,這與基于松散偏好規(guī)則圍捕時Leader涌現(xiàn)的判斷條件正好相反,其原因是簡化虛擬受力模型中機器人個體感知兩最近鄰的施力建模為斥力,Leader距離兩最近鄰越遠斥力越小,這樣會使得Leader涌現(xiàn)過程中Leader本身的振蕩較小。(2)未知非凸障礙物環(huán)境中非完整移動群機器人圍捕。提出了基于簡化虛擬受力模型的非凸靜態(tài)障礙物循障算法,該循障算法同樣只需要兩最近鄰和目標的位置信息即可,所設定的與障礙物之間的運動角度加大了與障礙物平行方向上的運動速度,保證了避障的同時快速循障。在理論上分析了系統(tǒng)的穩(wěn)定性,并給出了成功循障非凸靜態(tài)障礙物的條件。(3)未知動態(tài)變形障礙物環(huán)境中群機器人圍捕問題�；诤喕摂M受力模型設計了動態(tài)變形障礙物循障算法,該算法同樣只需要兩最近鄰和目標的位置信息即可實現(xiàn),所設定的與障礙物之間的運動角度兼顧了與障礙物垂直方向的速度大小和與障礙物平行方向上的速度大小,可以做到快速循障的同時與動態(tài)變形障礙物保持合適的距離,該循障算法還可適用于對機器人、非凸動態(tài)非變形障礙物、非凸靜態(tài)障礙物以及動態(tài)或靜態(tài)凸障礙物的循障。在理論上分析了系統(tǒng)的穩(wěn)定性,并給出了成功循障動態(tài)變形障礙物的條件和機器人數(shù)量擴展時圍捕隊形的特點。(4)未知動態(tài)復雜非凸障礙物環(huán)境中群機器人協(xié)同多層圍捕。將原來的簡化虛擬受力模型中機器人個體只能計算單層圍捕圓周上的受力擴展至可以計算任意層圍捕圓周上的受力,機器人在層與層之間的移動由其內(nèi)層或同層兩最近鄰的位置信息來確定,基于以上設計的多層圍捕算法實現(xiàn)了復雜環(huán)境下群機器人圍捕系統(tǒng)的高可擴展性、高可靠性和強避障能力。在理論上分析了多層圍捕系統(tǒng)的穩(wěn)定性。(5)未知動態(tài)凸障礙物環(huán)境中群機器人協(xié)同多目標圍捕。將原來的簡化虛擬受力模型中機器人個體對單個目標的受力分析擴展至對多目標中任意一個目標的受力分析,每個機器人的圍捕目標由動態(tài)多目標任務分配算法來確定,該算法只需要面向多目標中心方向180o范圍內(nèi)的兩最近鄰任務信息來實現(xiàn)即需要的信息量少,算法本身是分布式的并且簡單高效。在理論上分析了多目標圍捕系統(tǒng)的穩(wěn)定性,并給出了多目標任務分配時對機器人群體內(nèi)數(shù)量的要求。(6)自行設計了物理圍捕試驗平臺SRf H(Swarm Robots for Hunting,圍捕群機器人),驗證了所提基本圍捕理論與算法的正確性和有效性。采用LINK UWB定位系統(tǒng),開展了基于簡化虛擬受力模型的圍捕理論與算法在無障礙物環(huán)境下和凸障礙物環(huán)境下的圍捕試驗。不同環(huán)境和不同數(shù)目的SRf H圍捕試驗說明了所提出的圍捕理論與算法具有較好的魯棒性、可擴展性以及靈活性。Leader涌現(xiàn)的規(guī)律進一步在實踐中得到檢驗,由于通信環(huán)境的復雜性,再加上定位系統(tǒng)本身的定位誤差導致機器人個體關于兩最近鄰的受力分析難度進一步加大,除了有時候通過做濾波處理可以進行判斷外,其它情況下則需要將Leader的判斷條件更改為有一段受力序列是來自機器人個體左邊的兩最近鄰或右邊的兩最近鄰即可,不再需要單調衰減的限制,即實際情況中常出現(xiàn)單調但不一定衰減的情形,這是實踐與理想仿真中的不同之處。
[Abstract]:The research of group robot system is a hot spot recently. It is inspired by complex natural systems, such as social insects (ants, bees, etc.) or cooperative animal groups. The global behavior of the group robot system emerges from the local rules implemented on the level of the individual robot. The swarm robot capture system is a group robot system research. A very typical task platform, which has important research value, is conducive to the realization of encirclement, search and rescue, group confrontation, formation maintenance, cooperation handling, target guard and leadership guard, etc., which can be widely used in anti-terrorism, military, security and vigilance. However, the theory of the swarm robot capture system is not complete and involved. The trap environment is relatively simple, the expansibility of the capture system is not strong, and the multi target hunting algorithm is complex. In this paper, the swarm robot hunting system based on the artificial physical method is studied in this paper. The main contents are as follows: (1) the incompleteness of the mobile swarm robot in the unknown dynamic convex obstacle environment is different from the looseness based on looseness. This algorithm is based on the simplified virtual force model. This algorithm only needs two nearest neighbor and target location information. The algorithm is simple and efficient, and the simple number of force functions in the virtual force model is simplified. The bionic intelligent obstacle avoidance function realizes the good obstacle avoidance in the unknown dynamic convex obstacle environment. The stability of the trap system is analyzed in theory. The judgment condition of the emergence of the rule.Leader of Leader is the gradual approximate monotonous attenuation after the emergence of a large value, which is emerging from the emergence of the Leader based on the loose preference rule. The reason is the opposite. The reason is that the force modeling of the two nearest neighbor of the robot's individual perception in the simplified virtual force model is the repulsion, the farther the Leader distance two the nearest neighbor is, the smaller the repulsion, which will make the oscillation of the Leader itself less in the process of the emergence of the Leader. (2) the uncomplete mobile swarm robot in the unknown non convex obstacle environment is trapped. Based on the simplified virtual force model, a non convex static obstacle avoidance algorithm is proposed. The algorithm only needs the location information of the two nearest neighbor and the target. The motion angle between the obstacles and the obstacles is increased in parallel direction to the obstacle, and the obstacle avoidance is guaranteed at the same time. The stability of the system is analyzed, and the conditions for the successful non convex static obstacles are given. (3) the swarm robot capture in the unknown dynamic deformable obstacle environment. Based on the simplified virtual force model, the dynamic deformation obstacle algorithm is designed. The algorithm only needs the location information of the two nearest neighbour and the target. The motion angle between the obstacles and the obstacle takes into account the velocity of the vertical direction of the obstacle and the velocity in the parallel direction to the obstacle. It can keep the barrier and keep the appropriate distance to the dynamic deformable obstacle. The algorithm can also be applied to the robot, non convex dynamic non deformable obstacle, non convex static barrier. The obstacle and the obstacle of dynamic or static convex obstacle. The stability of the system is analyzed in theory, and the conditions of the obstacle dynamic deformable obstacle and the characteristic of the round formation when the number of robots are expanded. (4) the group robot in the unknown dynamic complex non convex obstacle environment is combined with the multi-layer trap. In the model, the robot can only calculate the force on the circumference of a single layer to calculate the force on the circumference of an arbitrary layer. The movement of the robot between the layer and the layer is determined by the location information of the nearest neighbor of the inner layer or the same layer two. High scalability, high reliability and strong obstacle avoidance ability. The stability of multi-layer capture system is analyzed theoretically. (5) group robots in unknown dynamic convex obstacle environment are cooperative multi target hunting. The force analysis of a single target in the original simplified virtual force model is extended to any one of the multiple targets. The target's force analysis is determined by the dynamic multi target assignment algorithm for each robot. The algorithm only needs the two nearest neighbor task information in the 180o range of the multi target center direction to realize the required information. The algorithm itself is distributed and simple and efficient. In theory, the multi target hunting system is analyzed. The stability of the system, and the requirements for the number of robot groups in the assignment of multi-objective tasks. (6) the physical capture test platform SRf H (Swarm Robots for Hunting, the swarm robot) is designed, and the correctness and effectiveness of the proposed basic capture theory and algorithm are verified. The LINK UWB positioning system is adopted and the simplification is carried out. The trap theory of virtual force model and its algorithm under the environment of barrier free environment and convex obstacle environment. The different environment and different number of SRf H capture experiments show that the proposed trap theory and algorithm have good robustness, scalability and flexibility of the emergence of.Leader are further tested in practice. As a result of the complexity of the communication environment, and the positioning error of the positioning system itself, the difficulty of the force analysis of the two nearest neighbor of the robot is further increased, except sometimes by the filtering process. In other cases, the judgment condition of the Leader is changed to a section of the force sequence. The nearest neighbor of the two nearest neighbor or the two nearest neighbor on the left of a robot is no longer required for monotonous attenuation, that is, a case of monotonous but not constant attenuation in the actual situation, which is the difference between practice and ideal simulation.

【學位授予單位】：湖南大學
【學位級別】：博士
【學位授予年份】：2016
【分類號】：TP242
，

本文編號：1849564