目标围捕任务中搜索与预包围阶段的群机器人行为学习

Swarm Robotic Behaviour Learning in Search and Pre-Surround Stage for Targets Trapping Task

为了进行群机器人协同作业,提出目标搜索中导航类集体行为学习策略.在使用具有闭环调节功能的动态任务分工方法进行任务分配、自组织地生成多个子群后,在子群中引入基于社会学习微粒群算法的机器人行为学习策略.在子群框架内,机器人各自独立地以感知的共同意向目标信号强度为标准对所有成员排序,将感知优于自己的机器人作为行为示范者.然后在搜索空间各维度上分别随机选择一个行为示范者,学习其在相应维度上的位置坐标,经构造得到搜索空间中自己的学习行为向量,由此决策自身的运动行为.仿真结果表明,在不需要学习全局社会经验的前提下,机器人能针对所属子群的共同意向目标进行协同作业,提高搜索效率.

A strategy for navigation-type collective behaviour learning is developed for swarm robotic coordination in a target search task. Sub-swarms are formed by utilizing the method of dynamic selforganizing task allocation with closed-loop adjusting function,and then a social learning particle swarm optimization based robotic learning strategy is introduced into sub-swarms. In the sub-swarm,all robots are sorted in descending order by the cognition ability of each robot to its common desired target. The robots with better perception of the target are regarded as the behaviour demonstrators. Then,one of the behaviour demonstrators is selected randomly by each robot to learn in every dimension of the working space. Thus,the learning behaviour vector of each robot can be constructed for decision making on its future moving behaviour. The results show that the robot can coordinate with each other and the search efficiency is improved without the global social experience learning.