开放环境中基于自适应分布控制的群体机器人多目标搜索方法  

Adaptive intra-group cooperative control based multiple-target search method for swarm robotics in boundless environment

作　　者：谭营[1,2,3] 陈人龙 翟天一 TAN Ying;CHEN RenLong;ZHAI TianYi(School of Artificial Intelligence,Peking University,Beijing 100871,China;Key Laboratory of Machine Perception and Intelligence(MOE),Beijing 100871,China;Institute for Artificial Intellignce,Peking University,Beijing 100871,China)

机构地区：[1]北京大学智能学院,北京100871 [2]机器感知与智能教育部重点实验室,北京100871 [3]北京大学人工智能研究院,北京100871

出　　处：《中国科学：技术科学》2023年第3期395-404,共10页Scientia Sinica(Technologica)

基　　金：国家重点研发计划(编号:2018AAA0102301);国家自然科学基金(批准号:62076010)资助项目。

摘　　要：多目标搜索问题是群体机器人一个重要的研究方向.现有工作多集中在带边界空间内的多目标搜索问题,而在开放环境中,探索机制会导致群体分散性过强而减弱探索能力.本文通过引入自适应扩散回归策略,在带有假目标的开放环境中,提出了具有高鲁棒性和适应性的群体机器人多目标搜索算法.文中首先从初始状态和处理假目标两方面对现有的主流群体机器人多目标搜索算法进行优化;基于机器人分布控制,本文对自适应群体机器人粒子群优化算法进行优化,提出基于自适应分布控制的群体机器人粒子群优化算法;其次,基于概率有限状态机搜索算法(PFSMS)对开放环境中的多目标搜索算法进行进一步的探索,本文以搜索时间为切入点,在PFSMS原有三种状态的基础上,添加回归状态作为附加状态,提出了基于自适应分布控制的概率有限状态机搜索算法(DPFSMS).当智能体的探索时间超过阈值时,智能体的速度由回归分量和扩散/搜索分量构成.DPFSMS算法给出了在无边界开放环境中的搜索策略,通过限制群体的扩散速度来自适应地调整智能体在无适应度值区域的运动随机性.最后,本文将DPFSMS算法与现有方法进行了对比,在对比实验中DPFSMS算法取得了目前最好的效果.The multi-objective search problem is one of the most important research aspects of swarm robots.The majority of existing research focuses on the multi-objective search problem in bounded space,whereas in the boundless environment,the existing work’s exploration mechanism leads to excessive group dispersion and weakens the exploration ability.A multi-objective search algorithm for swarm robots with high robustness and adaptability in a boundless environment with false targets is proposed by introducing an adaptive diffusion regression strategy.First,we optimize the existing multi-objective search algorithm of mainstream swarm robots from the aspects of the initial state and false target processing.Based on the distributed control of the robot,this study optimizes the adaptive swarm robot particle swarm optimization algorithm and proposes the distributed adaptive robot particle swarm optimization algorithm.Then,based on the Probabilistic Finite State Machine based Strategy(PFSMS)algorithm,this study investigates the multiobjective search algorithm in a boundless environment,adding the return state as an additional state to the original three states of PFSMS,using exploration time as the starting point to investigate effective search algorithms in a boundless environment,and proposes An Improved Simplified Three-State PFSM(DPFSMS).When the agent’s exploration time exceeds the threshold,the velocity of the agent is made up of regression and diffusion(search)components.The algorithm essentially provides a walking strategy in large blank areas of the boundless environment and reduces the randomness of the speed in this area by adaptively adjusting the agent group’s diffusion speed.Finally,DPFSMS is compared against existing Multiple-Target Search Algorithms;in our experiments,the DPFSMS algorithm has achieved the best effect in the comparative experiment.

关 键 词：群体机器人 多目标搜索 开放环境 假目标 群体智能 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置] TP18[自动化与计算机技术—控制科学与工程]