基于分层仿生神经网络的多机器人协同区域搜索算法  

A Hierarchical Bio-inspired Neural Network Based Multi-robot Cooperative Area Search Algorithm

作　　者：陈波 张辉[1,2] 江一鸣 钟杭 王耀南 CHEN Bo;ZHANG Hui;JIANG Yi-Ming;ZHONG Hang;WANG Yao-Nan(School of Robotics,Hunan University,Changsha 410082;National Engineering Research Center of Robot Visual Perception and Control Technology,Changsha 410082)

机构地区：[1]湖南大学机器人学院,长沙410082 [2]机器人视觉感知与控制技术国家工程研究中心,长沙410082

出　　处：《自动化学报》2025年第4期890-902,共13页Acta Automatica Sinica

基　　金：科技创新2030—“新一代人工智能”重大项目(2021ZD0114503);国家自然科学基金重大研究计划(92148204);湖南省科技创新领军人才(2022RC3063);湖南省十大技术攻关项目(2024GK1010);湖南省重点研发计划(2023GK2068,2022GK2011)资助。

摘　　要：针对多机器人系统在战场、灾难现场等复杂未知环境下的区域搜索问题,提出一种基于分层仿生神经网络的多机器人协同区域搜索算法.首先将仿生神经网络(Bio-inspired neural network,BNN)和不同分辨率下的区域栅格地图结合,构建分层仿生神经网络信息模型,其中包括区域搜索神经网络信息模型(Area search neural network information model,AS-BNN)和区域覆盖神经网络信息模型(Area coverage neural network information model,AC-BNN).机器人在任务区域内实时探测到的环境信息将转换为AS-BNN和AC-BNN中神经元的动态活性值.其次,在分层仿生神经网络信息模型基础上引入分布式模型预测控制(Distributed model predictive control,DMPC)框架,并设计多机器人分层协同决策机制.当机器人处于正常搜索状态时,基于AS-BNN进行搜索路径滚动优化决策;当机器人陷入局部最优状态时,则启用ACBNN引导机器人快速找到新的未搜索区域.最后,在复杂未知环境下进行多机器人区域搜索仿真实验,并与该领域内的3种算法进行比较.仿真结果验证了所提算法能够在复杂未知环境下引导多机器人系统高效地完成区域搜索任务.Aiming at the problem of multi-robot system area search in complex and unknown environments,such as battlefields and disaster scenes,a multi-robot cooperative area search algorithm based on a hierarchical bio-inspired neural network is proposed.Firstly,a hierarchical bio-inspired neural network(BNN)information model is constructed by combining the bio-inspired neural network with area grid maps at different resolutions,including the area search neural network information model(AS-BNN)and the area coverage neural network information model(ACBNN).The real-time environmental information detected by the robots in the task area is converted into the dynamic activity values of neurons in both AS-BNN and AC-BNN.Secondly,a distributed model predictive control(DMPC)framework is introduced based on the hierarchical bio-inspired neural network information model,and a multi-robot hierarchical cooperative decision-making mechanism is designed.When the robot is in a normal search state,a rolling optimization decision for the search path is made based on the AS-BNN.If the robot falls into a local optimum state,the AC-BNN is activated to guide it in quickly finding a new unsearched area.Finally,a multirobot area search simulation experiment is conducted in a complex and unknown environment,comparing the proposed algorithm with three other algorithms in this field.The simulation results verify that the proposed algorithm can guide the multi-robot system to efficiently complete the area search task in complex and unknown environments.

关 键 词：未知环境 多机器人系统 区域搜索 仿生神经网络 分布式模型预测控制 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]